Initial commit of source files

src/common_audio/signal_processing_library/OWNERS

@@ -0,0 +1,3 @@
+bjornv@webrtc.org
+tina.legrand@webrtc.org
+jan.skoglund@webrtc.org

src/common_audio/signal_processing_library/main/interface/spl_inl.h

@@ -0,0 +1,166 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+// This header file includes the inline functions in
+// the fix point signal processing library.
+
+#ifndef WEBRTC_SPL_SPL_INL_H_
+#define WEBRTC_SPL_SPL_INL_H_
+
+#ifdef WEBRTC_ARCH_ARM_V7A
+#include "spl_inl_armv7.h"
+#else
+
+static __inline WebRtc_Word16 WebRtcSpl_AddSatW16(WebRtc_Word16 a,
+                                                  WebRtc_Word16 b) {
+  WebRtc_Word32 s_sum = (WebRtc_Word32) a + (WebRtc_Word32) b;
+
+  if (s_sum > WEBRTC_SPL_WORD16_MAX)
+    s_sum = WEBRTC_SPL_WORD16_MAX;
+  else if (s_sum < WEBRTC_SPL_WORD16_MIN)
+    s_sum = WEBRTC_SPL_WORD16_MIN;
+
+  return (WebRtc_Word16)s_sum;
+}
+
+static __inline WebRtc_Word32 WebRtcSpl_AddSatW32(WebRtc_Word32 l_var1,
+                                                  WebRtc_Word32 l_var2) {
+  WebRtc_Word32 l_sum;
+
+  // perform long addition
+  l_sum = l_var1 + l_var2;
+
+  // check for under or overflow
+  if (WEBRTC_SPL_IS_NEG(l_var1)) {
+    if (WEBRTC_SPL_IS_NEG(l_var2) && !WEBRTC_SPL_IS_NEG(l_sum)) {
+        l_sum = (WebRtc_Word32)0x80000000;
+    }
+  } else {
+    if (!WEBRTC_SPL_IS_NEG(l_var2) && WEBRTC_SPL_IS_NEG(l_sum)) {
+        l_sum = (WebRtc_Word32)0x7FFFFFFF;
+    }
+  }
+
+  return l_sum;
+}
+
+static __inline WebRtc_Word16 WebRtcSpl_SubSatW16(WebRtc_Word16 var1,
+                                                  WebRtc_Word16 var2) {
+  WebRtc_Word32 l_diff;
+  WebRtc_Word16 s_diff;
+
+  // perform subtraction
+  l_diff = (WebRtc_Word32)var1 - (WebRtc_Word32)var2;
+
+  // default setting
+  s_diff = (WebRtc_Word16) l_diff;
+
+  // check for overflow
+  if (l_diff > (WebRtc_Word32)32767)
+  s_diff = (WebRtc_Word16)32767;
+
+  // check for underflow
+  if (l_diff < (WebRtc_Word32)-32768)
+  s_diff = (WebRtc_Word16)-32768;
+
+  return s_diff;
+}
+
+static __inline WebRtc_Word32 WebRtcSpl_SubSatW32(WebRtc_Word32 l_var1,
+                                                  WebRtc_Word32 l_var2) {
+  WebRtc_Word32 l_diff;
+
+  // perform subtraction
+  l_diff = l_var1 - l_var2;
+
+  // check for underflow
+  if ((l_var1 < 0) && (l_var2 > 0) && (l_diff > 0))
+    l_diff = (WebRtc_Word32)0x80000000;
+  // check for overflow
+  if ((l_var1 > 0) && (l_var2 < 0) && (l_diff < 0))
+    l_diff = (WebRtc_Word32)0x7FFFFFFF;
+
+  return l_diff;
+}
+
+static __inline WebRtc_Word16 WebRtcSpl_GetSizeInBits(WebRtc_UWord32 n) {
+  int bits;
+
+  if (0xFFFF0000 & n) {
+    bits = 16;
+  } else {
+    bits = 0;
+  }
+  if (0x0000FF00 & (n >> bits)) bits += 8;
+  if (0x000000F0 & (n >> bits)) bits += 4;
+  if (0x0000000C & (n >> bits)) bits += 2;
+  if (0x00000002 & (n >> bits)) bits += 1;
+  if (0x00000001 & (n >> bits)) bits += 1;
+
+  return bits;
+}
+
+static __inline int WebRtcSpl_NormW32(WebRtc_Word32 a) {
+  int zeros;
+
+  if (a <= 0) a ^= 0xFFFFFFFF;
+
+  if (!(0xFFFF8000 & a)) {
+    zeros = 16;
+  } else {
+    zeros = 0;
+  }
+  if (!(0xFF800000 & (a << zeros))) zeros += 8;
+  if (!(0xF8000000 & (a << zeros))) zeros += 4;
+  if (!(0xE0000000 & (a << zeros))) zeros += 2;
+  if (!(0xC0000000 & (a << zeros))) zeros += 1;
+
+  return zeros;
+}
+
+static __inline int WebRtcSpl_NormU32(WebRtc_UWord32 a) {
+  int zeros;
+
+  if (a == 0) return 0;
+
+  if (!(0xFFFF0000 & a)) {
+    zeros = 16;
+  } else {
+    zeros = 0;
+  }
+  if (!(0xFF000000 & (a << zeros))) zeros += 8;
+  if (!(0xF0000000 & (a << zeros))) zeros += 4;
+  if (!(0xC0000000 & (a << zeros))) zeros += 2;
+  if (!(0x80000000 & (a << zeros))) zeros += 1;
+
+  return zeros;
+}
+
+static __inline int WebRtcSpl_NormW16(WebRtc_Word16 a) {
+  int zeros;
+
+  if (a <= 0) a ^= 0xFFFF;
+
+  if (!(0xFF80 & a)) {
+    zeros = 8;
+  } else {
+    zeros = 0;
+  }
+  if (!(0xF800 & (a << zeros))) zeros += 4;
+  if (!(0xE000 & (a << zeros))) zeros += 2;
+  if (!(0xC000 & (a << zeros))) zeros += 1;
+
+  return zeros;
+}
+
+#endif  // WEBRTC_ARCH_ARM_V7A
+
+#endif  // WEBRTC_SPL_SPL_INL_H_

src/common_audio/signal_processing_library/main/interface/spl_inl_armv7.h

@@ -0,0 +1,122 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+// This header file includes the inline functions for ARM processors in
+// the fix point signal processing library.
+
+#ifndef WEBRTC_SPL_SPL_INL_ARMV7_H_
+#define WEBRTC_SPL_SPL_INL_ARMV7_H_
+
+static __inline WebRtc_Word32 WEBRTC_SPL_MUL_16_32_RSFT16(WebRtc_Word16 a,
+                                                          WebRtc_Word32 b) {
+  WebRtc_Word32 tmp;
+  __asm__("smulwb %0, %1, %2":"=r"(tmp):"r"(b), "r"(a));
+  return tmp;
+}
+
+static __inline WebRtc_Word32 WEBRTC_SPL_MUL_32_32_RSFT32(WebRtc_Word16 a,
+                                                          WebRtc_Word16 b,
+                                                          WebRtc_Word32 c) {
+  WebRtc_Word32 tmp;
+  __asm__("pkhbt %0, %1, %2, lsl #16" : "=r"(tmp) : "r"(b), "r"(a));
+  __asm__("smmul %0, %1, %2":"=r"(tmp):"r"(tmp), "r"(c));
+  return tmp;
+}
+
+static __inline WebRtc_Word32 WEBRTC_SPL_MUL_32_32_RSFT32BI(WebRtc_Word32 a,
+                                                            WebRtc_Word32 b) {
+  WebRtc_Word32 tmp;
+  __asm__("smmul %0, %1, %2":"=r"(tmp):"r"(a), "r"(b));
+  return tmp;
+}
+
+static __inline WebRtc_Word32 WEBRTC_SPL_MUL_16_16(WebRtc_Word16 a,
+                                                   WebRtc_Word16 b) {
+  WebRtc_Word32 tmp;
+  __asm__("smulbb %0, %1, %2":"=r"(tmp):"r"(a), "r"(b));
+  return tmp;
+}
+
+static __inline WebRtc_Word16 WebRtcSpl_AddSatW16(WebRtc_Word16 a,
+                                                  WebRtc_Word16 b) {
+  WebRtc_Word32 s_sum;
+
+  __asm__("qadd16 %0, %1, %2":"=r"(s_sum):"r"(a), "r"(b));
+
+  return (WebRtc_Word16) s_sum;
+}
+
+static __inline WebRtc_Word32 WebRtcSpl_AddSatW32(WebRtc_Word32 l_var1,
+                                                  WebRtc_Word32 l_var2) {
+  WebRtc_Word32 l_sum;
+
+  __asm__("qadd %0, %1, %2":"=r"(l_sum):"r"(l_var1), "r"(l_var2));
+
+  return l_sum;
+}
+
+static __inline WebRtc_Word16 WebRtcSpl_SubSatW16(WebRtc_Word16 var1,
+                                                  WebRtc_Word16 var2) {
+  WebRtc_Word32 s_sub;
+
+  __asm__("qsub16 %0, %1, %2":"=r"(s_sub):"r"(var1), "r"(var2));
+
+  return (WebRtc_Word16)s_sub;
+}
+
+static __inline WebRtc_Word32 WebRtcSpl_SubSatW32(WebRtc_Word32 l_var1,
+                                                  WebRtc_Word32 l_var2) {
+  WebRtc_Word32 l_sub;
+
+  __asm__("qsub %0, %1, %2":"=r"(l_sub):"r"(l_var1), "r"(l_var2));
+
+  return l_sub;
+}
+
+static __inline WebRtc_Word16 WebRtcSpl_GetSizeInBits(WebRtc_UWord32 n) {
+  WebRtc_Word32 tmp;
+
+  __asm__("clz %0, %1":"=r"(tmp):"r"(n));
+
+  return (WebRtc_Word16)(32 - tmp);
+}
+
+static __inline int WebRtcSpl_NormW32(WebRtc_Word32 a) {
+  WebRtc_Word32 tmp;
+
+  if (a <= 0) a ^= 0xFFFFFFFF;
+
+  __asm__("clz %0, %1":"=r"(tmp):"r"(a));
+
+  return tmp - 1;
+}
+
+static __inline int WebRtcSpl_NormU32(WebRtc_UWord32 a) {
+  int tmp;
+
+  if (a == 0) return 0;
+
+  __asm__("clz %0, %1":"=r"(tmp):"r"(a));
+
+  return tmp;
+}
+
+static __inline int WebRtcSpl_NormW16(WebRtc_Word16 a) {
+  WebRtc_Word32 tmp;
+
+  if (a <= 0) a ^= 0xFFFFFFFF;
+
+  __asm__("clz %0, %1":"=r"(tmp):"r"(a));
+
+  return tmp - 17;
+}
+
+#endif  // WEBRTC_SPL_SPL_INL_ARMV7_H_

src/common_audio/signal_processing_library/main/source/auto_corr_to_refl_coef.c

@@ -0,0 +1,103 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_AutoCorrToReflCoef().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+void WebRtcSpl_AutoCorrToReflCoef(G_CONST WebRtc_Word32 *R, int use_order, WebRtc_Word16 *K)
+{
+    int i, n;
+    WebRtc_Word16 tmp;
+    G_CONST WebRtc_Word32 *rptr;
+    WebRtc_Word32 L_num, L_den;
+    WebRtc_Word16 *acfptr, *pptr, *wptr, *p1ptr, *w1ptr, ACF[WEBRTC_SPL_MAX_LPC_ORDER],
+            P[WEBRTC_SPL_MAX_LPC_ORDER], W[WEBRTC_SPL_MAX_LPC_ORDER];
+
+    // Initialize loop and pointers.
+    acfptr = ACF;
+    rptr = R;
+    pptr = P;
+    p1ptr = &P[1];
+    w1ptr = &W[1];
+    wptr = w1ptr;
+
+    // First loop; n=0. Determine shifting.
+    tmp = WebRtcSpl_NormW32(*R);
+    *acfptr = (WebRtc_Word16)((*rptr++ << tmp) >> 16);
+    *pptr++ = *acfptr++;
+
+    // Initialize ACF, P and W.
+    for (i = 1; i <= use_order; i++)
+    {
+        *acfptr = (WebRtc_Word16)((*rptr++ << tmp) >> 16);
+        *wptr++ = *acfptr;
+        *pptr++ = *acfptr++;
+    }
+
+    // Compute reflection coefficients.
+    for (n = 1; n <= use_order; n++, K++)
+    {
+        tmp = WEBRTC_SPL_ABS_W16(*p1ptr);
+        if (*P < tmp)
+        {
+            for (i = n; i <= use_order; i++)
+                *K++ = 0;
+
+            return;
+        }
+
+        // Division: WebRtcSpl_div(tmp, *P)
+        *K = 0;
+        if (tmp != 0)
+        {
+            L_num = tmp;
+            L_den = *P;
+            i = 15;
+            while (i--)
+            {
+                (*K) <<= 1;
+                L_num <<= 1;
+                if (L_num >= L_den)
+                {
+                    L_num -= L_den;
+                    (*K)++;
+                }
+            }
+            if (*p1ptr > 0)
+                *K = -*K;
+        }
+
+        // Last iteration; don't do Schur recursion.
+        if (n == use_order)
+            return;
+
+        // Schur recursion.
+        pptr = P;
+        wptr = w1ptr;
+        tmp = (WebRtc_Word16)(((WebRtc_Word32)*p1ptr * (WebRtc_Word32)*K + 16384) >> 15);
+        *pptr = WEBRTC_SPL_ADD_SAT_W16( *pptr, tmp );
+        pptr++;
+        for (i = 1; i <= use_order - n; i++)
+        {
+            tmp = (WebRtc_Word16)(((WebRtc_Word32)*wptr * (WebRtc_Word32)*K + 16384) >> 15);
+            *pptr = WEBRTC_SPL_ADD_SAT_W16( *(pptr+1), tmp );
+            pptr++;
+            tmp = (WebRtc_Word16)(((WebRtc_Word32)*pptr * (WebRtc_Word32)*K + 16384) >> 15);
+            *wptr = WEBRTC_SPL_ADD_SAT_W16( *wptr, tmp );
+            wptr++;
+        }
+    }
+}

src/common_audio/signal_processing_library/main/source/auto_correlation.c

@@ -0,0 +1,141 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_AutoCorrelation().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+int WebRtcSpl_AutoCorrelation(G_CONST WebRtc_Word16* in_vector,
+                              int in_vector_length,
+                              int order,
+                              WebRtc_Word32* result,
+                              int* scale)
+{
+    WebRtc_Word32 sum;
+    int i, j;
+    WebRtc_Word16 smax; // Sample max
+    G_CONST WebRtc_Word16* xptr1;
+    G_CONST WebRtc_Word16* xptr2;
+    WebRtc_Word32* resultptr;
+    int scaling = 0;
+
+#ifdef _ARM_OPT_
+#pragma message("NOTE: _ARM_OPT_ optimizations are used")
+    WebRtc_Word16 loops4;
+#endif
+
+    if (order < 0)
+        order = in_vector_length;
+
+    // Find the max. sample
+    smax = WebRtcSpl_MaxAbsValueW16(in_vector, in_vector_length);
+
+    // In order to avoid overflow when computing the sum we should scale the samples so that
+    // (in_vector_length * smax * smax) will not overflow.
+
+    if (smax == 0)
+    {
+        scaling = 0;
+    } else
+    {
+        int nbits = WebRtcSpl_GetSizeInBits(in_vector_length); // # of bits in the sum loop
+        int t = WebRtcSpl_NormW32(WEBRTC_SPL_MUL(smax, smax)); // # of bits to normalize smax
+
+        if (t > nbits)
+        {
+            scaling = 0;
+        } else
+        {
+            scaling = nbits - t;
+        }
+
+    }
+
+    resultptr = result;
+
+    // Perform the actual correlation calculation
+    for (i = 0; i < order + 1; i++)
+    {
+        int loops = (in_vector_length - i);
+        sum = 0;
+        xptr1 = in_vector;
+        xptr2 = &in_vector[i];
+#ifndef _ARM_OPT_
+        for (j = loops; j > 0; j--)
+        {
+            sum += WEBRTC_SPL_MUL_16_16_RSFT(*xptr1++, *xptr2++, scaling);
+        }
+#else
+        loops4 = (loops >> 2) << 2;
+
+        if (scaling == 0)
+        {
+            for (j = 0; j < loops4; j = j + 4)
+            {
+                sum += WEBRTC_SPL_MUL_16_16(*xptr1, *xptr2);
+                xptr1++;
+                xptr2++;
+                sum += WEBRTC_SPL_MUL_16_16(*xptr1, *xptr2);
+                xptr1++;
+                xptr2++;
+                sum += WEBRTC_SPL_MUL_16_16(*xptr1, *xptr2);
+                xptr1++;
+                xptr2++;
+                sum += WEBRTC_SPL_MUL_16_16(*xptr1, *xptr2);
+                xptr1++;
+                xptr2++;
+            }
+
+            for (j = loops4; j < loops; j++)
+            {
+                sum += WEBRTC_SPL_MUL_16_16(*xptr1, *xptr2);
+                xptr1++;
+                xptr2++;
+            }
+        }
+        else
+        {
+            for (j = 0; j < loops4; j = j + 4)
+            {
+                sum += WEBRTC_SPL_MUL_16_16_RSFT(*xptr1, *xptr2, scaling);
+                xptr1++;
+                xptr2++;
+                sum += WEBRTC_SPL_MUL_16_16_RSFT(*xptr1, *xptr2, scaling);
+                xptr1++;
+                xptr2++;
+                sum += WEBRTC_SPL_MUL_16_16_RSFT(*xptr1, *xptr2, scaling);
+                xptr1++;
+                xptr2++;
+                sum += WEBRTC_SPL_MUL_16_16_RSFT(*xptr1, *xptr2, scaling);
+                xptr1++;
+                xptr2++;
+            }
+
+            for (j = loops4; j < loops; j++)
+            {
+                sum += WEBRTC_SPL_MUL_16_16_RSFT(*xptr1, *xptr2, scaling);
+                xptr1++;
+                xptr2++;
+            }
+        }
+
+#endif
+        *resultptr++ = sum;
+    }
+
+    *scale = scaling;
+
+    return order + 1;
+}

src/common_audio/signal_processing_library/main/source/complex_bit_reverse.c

@@ -0,0 +1,51 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_ComplexBitReverse().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+void WebRtcSpl_ComplexBitReverse(WebRtc_Word16 frfi[], int stages)
+{
+    int mr, nn, n, l, m;
+    WebRtc_Word16 tr, ti;
+
+    n = 1 << stages;
+
+    mr = 0;
+    nn = n - 1;
+
+    // decimation in time - re-order data
+    for (m = 1; m <= nn; ++m)
+    {
+        l = n;
+        do
+        {
+            l >>= 1;
+        } while (mr + l > nn);
+        mr = (mr & (l - 1)) + l;
+
+        if (mr <= m)
+            continue;
+
+        tr = frfi[2 * m];
+        frfi[2 * m] = frfi[2 * mr];
+        frfi[2 * mr] = tr;
+
+        ti = frfi[2 * m + 1];
+        frfi[2 * m + 1] = frfi[2 * mr + 1];
+        frfi[2 * mr + 1] = ti;
+    }
+}

src/common_audio/signal_processing_library/main/source/complex_fft.c

@@ -0,0 +1,150 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_ComplexFFT().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+#define CFFTSFT 14
+#define CFFTRND 1
+#define CFFTRND2 16384
+
+int WebRtcSpl_ComplexFFT(WebRtc_Word16 frfi[], int stages, int mode)
+{
+    int i, j, l, k, istep, n, m;
+    WebRtc_Word16 wr, wi;
+    WebRtc_Word32 tr32, ti32, qr32, qi32;
+
+    /* The 1024-value is a constant given from the size of WebRtcSpl_kSinTable1024[],
+     * and should not be changed depending on the input parameter 'stages'
+     */
+    n = 1 << stages;
+    if (n > 1024)
+        return -1;
+
+    l = 1;
+    k = 10 - 1; /* Constant for given WebRtcSpl_kSinTable1024[]. Do not change
+         depending on the input parameter 'stages' */
+
+    if (mode == 0)
+    {
+        // mode==0: Low-complexity and Low-accuracy mode
+        while (l < n)
+        {
+            istep = l << 1;
+
+            for (m = 0; m < l; ++m)
+            {
+                j = m << k;
+
+                /* The 256-value is a constant given as 1/4 of the size of
+                 * WebRtcSpl_kSinTable1024[], and should not be changed depending on the input
+                 * parameter 'stages'. It will result in 0 <= j < N_SINE_WAVE/2
+                 */
+                wr = WebRtcSpl_kSinTable1024[j + 256];
+                wi = -WebRtcSpl_kSinTable1024[j];
+
+                for (i = m; i < n; i += istep)
+                {
+                    j = i + l;
+
+                    tr32 = WEBRTC_SPL_RSHIFT_W32((WEBRTC_SPL_MUL_16_16(wr, frfi[2 * j])
+                            - WEBRTC_SPL_MUL_16_16(wi, frfi[2 * j + 1])), 15);
+
+                    ti32 = WEBRTC_SPL_RSHIFT_W32((WEBRTC_SPL_MUL_16_16(wr, frfi[2 * j + 1])
+                            + WEBRTC_SPL_MUL_16_16(wi, frfi[2 * j])), 15);
+
+                    qr32 = (WebRtc_Word32)frfi[2 * i];
+                    qi32 = (WebRtc_Word32)frfi[2 * i + 1];
+                    frfi[2 * j] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(qr32 - tr32, 1);
+                    frfi[2 * j + 1] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(qi32 - ti32, 1);
+                    frfi[2 * i] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(qr32 + tr32, 1);
+                    frfi[2 * i + 1] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(qi32 + ti32, 1);
+                }
+            }
+
+            --k;
+            l = istep;
+
+        }
+
+    } else
+    {
+        // mode==1: High-complexity and High-accuracy mode
+        while (l < n)
+        {
+            istep = l << 1;
+
+            for (m = 0; m < l; ++m)
+            {
+                j = m << k;
+
+                /* The 256-value is a constant given as 1/4 of the size of
+                 * WebRtcSpl_kSinTable1024[], and should not be changed depending on the input
+                 * parameter 'stages'. It will result in 0 <= j < N_SINE_WAVE/2
+                 */
+                wr = WebRtcSpl_kSinTable1024[j + 256];
+                wi = -WebRtcSpl_kSinTable1024[j];
+
+#ifdef WEBRTC_ARCH_ARM_V7A
+                WebRtc_Word32 wri;
+                WebRtc_Word32 frfi_r;
+                __asm__("pkhbt %0, %1, %2, lsl #16" : "=r"(wri) :
+                    "r"((WebRtc_Word32)wr), "r"((WebRtc_Word32)wi));
+#endif
+
+                for (i = m; i < n; i += istep)
+                {
+                    j = i + l;
+
+#ifdef WEBRTC_ARCH_ARM_V7A
+                    __asm__("pkhbt %0, %1, %2, lsl #16" : "=r"(frfi_r) :
+                        "r"((WebRtc_Word32)frfi[2*j]), "r"((WebRtc_Word32)frfi[2*j +1]));
+                    __asm__("smlsd %0, %1, %2, %3" : "=r"(tr32) :
+                        "r"(wri), "r"(frfi_r), "r"(CFFTRND));
+                    __asm__("smladx %0, %1, %2, %3" : "=r"(ti32) :
+                        "r"(wri), "r"(frfi_r), "r"(CFFTRND));
+    
+#else
+                    tr32 = WEBRTC_SPL_MUL_16_16(wr, frfi[2 * j])
+                            - WEBRTC_SPL_MUL_16_16(wi, frfi[2 * j + 1]) + CFFTRND;
+
+                    ti32 = WEBRTC_SPL_MUL_16_16(wr, frfi[2 * j + 1])
+                            + WEBRTC_SPL_MUL_16_16(wi, frfi[2 * j]) + CFFTRND;
+#endif
+
+                    tr32 = WEBRTC_SPL_RSHIFT_W32(tr32, 15 - CFFTSFT);
+                    ti32 = WEBRTC_SPL_RSHIFT_W32(ti32, 15 - CFFTSFT);
+
+                    qr32 = ((WebRtc_Word32)frfi[2 * i]) << CFFTSFT;
+                    qi32 = ((WebRtc_Word32)frfi[2 * i + 1]) << CFFTSFT;
+
+                    frfi[2 * j] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(
+                            (qr32 - tr32 + CFFTRND2), 1 + CFFTSFT);
+                    frfi[2 * j + 1] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(
+                            (qi32 - ti32 + CFFTRND2), 1 + CFFTSFT);
+                    frfi[2 * i] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(
+                            (qr32 + tr32 + CFFTRND2), 1 + CFFTSFT);
+                    frfi[2 * i + 1] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(
+                            (qi32 + ti32 + CFFTRND2), 1 + CFFTSFT);
+                }
+            }
+
+            --k;
+            l = istep;
+        }
+    }
+    return 0;
+}

src/common_audio/signal_processing_library/main/source/complex_ifft.c

@@ -0,0 +1,161 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_ComplexIFFT().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+#define CIFFTSFT 14
+#define CIFFTRND 1
+
+int WebRtcSpl_ComplexIFFT(WebRtc_Word16 frfi[], int stages, int mode)
+{
+    int i, j, l, k, istep, n, m, scale, shift;
+    WebRtc_Word16 wr, wi;
+    WebRtc_Word32 tr32, ti32, qr32, qi32;
+    WebRtc_Word32 tmp32, round2;
+
+    /* The 1024-value is a constant given from the size of WebRtcSpl_kSinTable1024[],
+     * and should not be changed depending on the input parameter 'stages'
+     */
+    n = 1 << stages;
+    if (n > 1024)
+        return -1;
+
+    scale = 0;
+
+    l = 1;
+    k = 10 - 1; /* Constant for given WebRtcSpl_kSinTable1024[]. Do not change
+         depending on the input parameter 'stages' */
+
+    while (l < n)
+    {
+        // variable scaling, depending upon data
+        shift = 0;
+        round2 = 8192;
+
+        tmp32 = (WebRtc_Word32)WebRtcSpl_MaxAbsValueW16(frfi, 2 * n);
+        if (tmp32 > 13573)
+        {
+            shift++;
+            scale++;
+            round2 <<= 1;
+        }
+        if (tmp32 > 27146)
+        {
+            shift++;
+            scale++;
+            round2 <<= 1;
+        }
+
+        istep = l << 1;
+
+        if (mode == 0)
+        {
+            // mode==0: Low-complexity and Low-accuracy mode
+            for (m = 0; m < l; ++m)
+            {
+                j = m << k;
+
+                /* The 256-value is a constant given as 1/4 of the size of
+                 * WebRtcSpl_kSinTable1024[], and should not be changed depending on the input
+                 * parameter 'stages'. It will result in 0 <= j < N_SINE_WAVE/2
+                 */
+                wr = WebRtcSpl_kSinTable1024[j + 256];
+                wi = WebRtcSpl_kSinTable1024[j];
+
+                for (i = m; i < n; i += istep)
+                {
+                    j = i + l;
+
+                    tr32 = WEBRTC_SPL_RSHIFT_W32((WEBRTC_SPL_MUL_16_16_RSFT(wr, frfi[2 * j], 0)
+                            - WEBRTC_SPL_MUL_16_16_RSFT(wi, frfi[2 * j + 1], 0)), 15);
+
+                    ti32 = WEBRTC_SPL_RSHIFT_W32(
+                            (WEBRTC_SPL_MUL_16_16_RSFT(wr, frfi[2 * j + 1], 0)
+                                    + WEBRTC_SPL_MUL_16_16_RSFT(wi,frfi[2*j],0)), 15);
+
+                    qr32 = (WebRtc_Word32)frfi[2 * i];
+                    qi32 = (WebRtc_Word32)frfi[2 * i + 1];
+                    frfi[2 * j] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(qr32 - tr32, shift);
+                    frfi[2 * j + 1] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(qi32 - ti32, shift);
+                    frfi[2 * i] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(qr32 + tr32, shift);
+                    frfi[2 * i + 1] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(qi32 + ti32, shift);
+                }
+            }
+        } else
+        {
+            // mode==1: High-complexity and High-accuracy mode
+
+            for (m = 0; m < l; ++m)
+            {
+                j = m << k;
+
+                /* The 256-value is a constant given as 1/4 of the size of
+                 * WebRtcSpl_kSinTable1024[], and should not be changed depending on the input
+                 * parameter 'stages'. It will result in 0 <= j < N_SINE_WAVE/2
+                 */
+                wr = WebRtcSpl_kSinTable1024[j + 256];
+                wi = WebRtcSpl_kSinTable1024[j];
+
+#ifdef WEBRTC_ARCH_ARM_V7A
+                WebRtc_Word32 wri;
+                WebRtc_Word32 frfi_r;
+                __asm__("pkhbt %0, %1, %2, lsl #16" : "=r"(wri) :
+                    "r"((WebRtc_Word32)wr), "r"((WebRtc_Word32)wi));
+#endif
+
+                for (i = m; i < n; i += istep)
+                {
+                    j = i + l;
+
+#ifdef WEBRTC_ARCH_ARM_V7A
+                    __asm__("pkhbt %0, %1, %2, lsl #16" : "=r"(frfi_r) :
+                        "r"((WebRtc_Word32)frfi[2*j]), "r"((WebRtc_Word32)frfi[2*j +1]));
+                    __asm__("smlsd %0, %1, %2, %3" : "=r"(tr32) :
+                        "r"(wri), "r"(frfi_r), "r"(CIFFTRND));
+                    __asm__("smladx %0, %1, %2, %3" : "=r"(ti32) :
+                        "r"(wri), "r"(frfi_r), "r"(CIFFTRND));
+#else
+
+                    tr32 = WEBRTC_SPL_MUL_16_16(wr, frfi[2 * j])
+                            - WEBRTC_SPL_MUL_16_16(wi, frfi[2 * j + 1]) + CIFFTRND;
+
+                    ti32 = WEBRTC_SPL_MUL_16_16(wr, frfi[2 * j + 1])
+                            + WEBRTC_SPL_MUL_16_16(wi, frfi[2 * j]) + CIFFTRND;
+#endif
+                    tr32 = WEBRTC_SPL_RSHIFT_W32(tr32, 15 - CIFFTSFT);
+                    ti32 = WEBRTC_SPL_RSHIFT_W32(ti32, 15 - CIFFTSFT);
+    
+                    qr32 = ((WebRtc_Word32)frfi[2 * i]) << CIFFTSFT;
+                    qi32 = ((WebRtc_Word32)frfi[2 * i + 1]) << CIFFTSFT;
+    
+                    frfi[2 * j] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32((qr32 - tr32+round2),
+                                                                       shift+CIFFTSFT);
+                    frfi[2 * j + 1] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(
+                            (qi32 - ti32 + round2), shift + CIFFTSFT);
+                    frfi[2 * i] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32((qr32 + tr32 + round2),
+                                                                       shift + CIFFTSFT);
+                    frfi[2 * i + 1] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(
+                            (qi32 + ti32 + round2), shift + CIFFTSFT);
+                }
+            }
+
+        }
+        --k;
+        l = istep;
+    }
+    return scale;
+}

src/common_audio/signal_processing_library/main/source/copy_set_operations.c

@@ -0,0 +1,108 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the implementation of functions
+ * WebRtcSpl_MemSetW16()
+ * WebRtcSpl_MemSetW32()
+ * WebRtcSpl_MemCpyReversedOrder()
+ * WebRtcSpl_CopyFromEndW16()
+ * WebRtcSpl_ZerosArrayW16()
+ * WebRtcSpl_ZerosArrayW32()
+ * WebRtcSpl_OnesArrayW16()
+ * WebRtcSpl_OnesArrayW32()
+ *
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include <string.h>
+#include "signal_processing_library.h"
+
+
+void WebRtcSpl_MemSetW16(WebRtc_Word16 *ptr, WebRtc_Word16 set_value, int length)
+{
+    int j;
+    WebRtc_Word16 *arrptr = ptr;
+
+    for (j = length; j > 0; j--)
+    {
+        *arrptr++ = set_value;
+    }
+}
+
+void WebRtcSpl_MemSetW32(WebRtc_Word32 *ptr, WebRtc_Word32 set_value, int length)
+{
+    int j;
+    WebRtc_Word32 *arrptr = ptr;
+
+    for (j = length; j > 0; j--)
+    {
+        *arrptr++ = set_value;
+    }
+}
+
+void WebRtcSpl_MemCpyReversedOrder(WebRtc_Word16* dest, WebRtc_Word16* source, int length)
+{
+    int j;
+    WebRtc_Word16* destPtr = dest;
+    WebRtc_Word16* sourcePtr = source;
+
+    for (j = 0; j < length; j++)
+    {
+        *destPtr-- = *sourcePtr++;
+    }
+}
+
+WebRtc_Word16 WebRtcSpl_CopyFromEndW16(G_CONST WebRtc_Word16 *vector_in,
+                                       WebRtc_Word16 length,
+                                       WebRtc_Word16 samples,
+                                       WebRtc_Word16 *vector_out)
+{
+    // Copy the last <samples> of the input vector to vector_out
+    WEBRTC_SPL_MEMCPY_W16(vector_out, &vector_in[length - samples], samples);
+
+    return samples;
+}
+
+WebRtc_Word16 WebRtcSpl_ZerosArrayW16(WebRtc_Word16 *vector, WebRtc_Word16 length)
+{
+    WebRtcSpl_MemSetW16(vector, 0, length);
+    return length;
+}
+
+WebRtc_Word16 WebRtcSpl_ZerosArrayW32(WebRtc_Word32 *vector, WebRtc_Word16 length)
+{
+    WebRtcSpl_MemSetW32(vector, 0, length);
+    return length;
+}
+
+WebRtc_Word16 WebRtcSpl_OnesArrayW16(WebRtc_Word16 *vector, WebRtc_Word16 length)
+{
+    WebRtc_Word16 i;
+    WebRtc_Word16 *tmpvec = vector;
+    for (i = 0; i < length; i++)
+    {
+        *tmpvec++ = 1;
+    }
+    return length;
+}
+
+WebRtc_Word16 WebRtcSpl_OnesArrayW32(WebRtc_Word32 *vector, WebRtc_Word16 length)
+{
+    WebRtc_Word16 i;
+    WebRtc_Word32 *tmpvec = vector;
+    for (i = 0; i < length; i++)
+    {
+        *tmpvec++ = 1;
+    }
+    return length;
+}

src/common_audio/signal_processing_library/main/source/cos_table.c

@@ -0,0 +1,60 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the 360 degree cos table.
+ *
+ */
+
+#include "signal_processing_library.h"
+
+WebRtc_Word16 WebRtcSpl_kCosTable[] = {
+        8192,  8190,  8187,  8180,  8172,  8160,  8147,  8130,  8112,
+        8091,  8067,  8041,  8012,  7982,  7948,  7912,  7874,  7834,
+        7791,  7745,  7697,  7647,  7595,  7540,  7483,  7424,  7362,
+        7299,  7233,  7164,  7094,  7021,  6947,  6870,  6791,  6710,
+        6627,  6542,  6455,  6366,  6275,  6182,  6087,  5991,  5892,
+        5792,  5690,  5586,  5481,  5374,  5265,  5155,  5043,  4930,
+        4815,  4698,  4580,  4461,  4341,  4219,  4096,  3971,  3845,
+        3719,  3591,  3462,  3331,  3200,  3068,  2935,  2801,  2667,
+        2531,  2395,  2258,  2120,  1981,  1842,  1703,  1563,  1422,
+        1281,  1140,   998,   856,   713,   571,   428,   285,   142,
+           0,  -142,  -285,  -428,  -571,  -713,  -856,  -998, -1140,
+       -1281, -1422, -1563, -1703, -1842, -1981, -2120, -2258, -2395,
+       -2531, -2667, -2801, -2935, -3068, -3200, -3331, -3462, -3591,
+       -3719, -3845, -3971, -4095, -4219, -4341, -4461, -4580, -4698,
+       -4815, -4930, -5043, -5155, -5265, -5374, -5481, -5586, -5690,
+       -5792, -5892, -5991, -6087, -6182, -6275, -6366, -6455, -6542,
+       -6627, -6710, -6791, -6870, -6947, -7021, -7094, -7164, -7233,
+       -7299, -7362, -7424, -7483, -7540, -7595, -7647, -7697, -7745,
+       -7791, -7834, -7874, -7912, -7948, -7982, -8012, -8041, -8067,
+       -8091, -8112, -8130, -8147, -8160, -8172, -8180, -8187, -8190,
+       -8191, -8190, -8187, -8180, -8172, -8160, -8147, -8130, -8112,
+       -8091, -8067, -8041, -8012, -7982, -7948, -7912, -7874, -7834,
+       -7791, -7745, -7697, -7647, -7595, -7540, -7483, -7424, -7362,
+       -7299, -7233, -7164, -7094, -7021, -6947, -6870, -6791, -6710,
+       -6627, -6542, -6455, -6366, -6275, -6182, -6087, -5991, -5892,
+       -5792, -5690, -5586, -5481, -5374, -5265, -5155, -5043, -4930,
+       -4815, -4698, -4580, -4461, -4341, -4219, -4096, -3971, -3845,
+       -3719, -3591, -3462, -3331, -3200, -3068, -2935, -2801, -2667,
+       -2531, -2395, -2258, -2120, -1981, -1842, -1703, -1563, -1422,
+       -1281, -1140,  -998,  -856,  -713,  -571,  -428,  -285,  -142,
+           0,   142,   285,   428,   571,   713,   856,   998,  1140,
+        1281,  1422,  1563,  1703,  1842,  1981,  2120,  2258,  2395,
+        2531,  2667,  2801,  2935,  3068,  3200,  3331,  3462,  3591,
+        3719,  3845,  3971,  4095,  4219,  4341,  4461,  4580,  4698,
+        4815,  4930,  5043,  5155,  5265,  5374,  5481,  5586,  5690,
+        5792,  5892,  5991,  6087,  6182,  6275,  6366,  6455,  6542,
+        6627,  6710,  6791,  6870,  6947,  7021,  7094,  7164,  7233,
+        7299,  7362,  7424,  7483,  7540,  7595,  7647,  7697,  7745,
+        7791,  7834,  7874,  7912,  7948,  7982,  8012,  8041,  8067,
+        8091,  8112,  8130,  8147,  8160,  8172,  8180,  8187,  8190
+};

src/common_audio/signal_processing_library/main/source/cross_correlation.c

@@ -0,0 +1,267 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_CrossCorrelation().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+void WebRtcSpl_CrossCorrelation(WebRtc_Word32* cross_correlation, WebRtc_Word16* seq1,
+                                WebRtc_Word16* seq2, WebRtc_Word16 dim_seq,
+                                WebRtc_Word16 dim_cross_correlation,
+                                WebRtc_Word16 right_shifts,
+                                WebRtc_Word16 step_seq2)
+{
+    int i, j;
+    WebRtc_Word16* seq1Ptr;
+    WebRtc_Word16* seq2Ptr;
+    WebRtc_Word32* CrossCorrPtr;
+
+#ifdef _XSCALE_OPT_
+
+#ifdef _WIN32
+#pragma message("NOTE: _XSCALE_OPT_ optimizations are used (overrides _ARM_OPT_ and requires /QRxscale compiler flag)")
+#endif
+
+    __int64 macc40;
+
+    int iseq1[250];
+    int iseq2[250];
+    int iseq3[250];
+    int * iseq1Ptr;
+    int * iseq2Ptr;
+    int * iseq3Ptr;
+    int len, i_len;
+
+    seq1Ptr = seq1;
+    iseq1Ptr = iseq1;
+    for(i = 0; i < ((dim_seq + 1) >> 1); i++)
+    {
+        *iseq1Ptr = (unsigned short)*seq1Ptr++;
+        *iseq1Ptr++ |= (WebRtc_Word32)*seq1Ptr++ << 16;
+
+    }
+
+    if(dim_seq%2)
+    {
+        *(iseq1Ptr-1) &= 0x0000ffff;
+    }
+    *iseq1Ptr = 0;
+    iseq1Ptr++;
+    *iseq1Ptr = 0;
+    iseq1Ptr++;
+    *iseq1Ptr = 0;
+
+    if(step_seq2 < 0)
+    {
+        seq2Ptr = seq2 - dim_cross_correlation + 1;
+        CrossCorrPtr = &cross_correlation[dim_cross_correlation - 1];
+    }
+    else
+    {
+        seq2Ptr = seq2;
+        CrossCorrPtr = cross_correlation;
+    }
+
+    len = dim_seq + dim_cross_correlation - 1;
+    i_len = (len + 1) >> 1;
+    iseq2Ptr = iseq2;
+
+    iseq3Ptr = iseq3;
+    for(i = 0; i < i_len; i++)
+    {
+        *iseq2Ptr = (unsigned short)*seq2Ptr++;
+        *iseq3Ptr = (unsigned short)*seq2Ptr;
+        *iseq2Ptr++ |= (WebRtc_Word32)*seq2Ptr++ << 16;
+        *iseq3Ptr++ |= (WebRtc_Word32)*seq2Ptr << 16;
+    }
+
+    if(len % 2)
+    {
+        iseq2[i_len - 1] &= 0x0000ffff;
+        iseq3[i_len - 1] = 0;
+    }
+    else
+    iseq3[i_len - 1] &= 0x0000ffff;
+
+    iseq2[i_len] = 0;
+    iseq3[i_len] = 0;
+    iseq2[i_len + 1] = 0;
+    iseq3[i_len + 1] = 0;
+    iseq2[i_len + 2] = 0;
+    iseq3[i_len + 2] = 0;
+
+    // Set pointer to start value
+    iseq2Ptr = iseq2;
+    iseq3Ptr = iseq3;
+
+    i_len = (dim_seq + 7) >> 3;
+    for (i = 0; i < dim_cross_correlation; i++)
+    {
+
+        iseq1Ptr = iseq1;
+
+        macc40 = 0;
+
+        _WriteCoProcessor(macc40, 0);
+
+        if((i & 1))
+        {
+            iseq3Ptr = iseq3 + (i >> 1);
+            for (j = i_len; j > 0; j--)
+            {
+                _SmulAddPack_2SW_ACC(*iseq1Ptr++, *iseq3Ptr++);
+                _SmulAddPack_2SW_ACC(*iseq1Ptr++, *iseq3Ptr++);
+                _SmulAddPack_2SW_ACC(*iseq1Ptr++, *iseq3Ptr++);
+                _SmulAddPack_2SW_ACC(*iseq1Ptr++, *iseq3Ptr++);
+            }
+        }
+        else
+        {
+            iseq2Ptr = iseq2 + (i >> 1);
+            for (j = i_len; j > 0; j--)
+            {
+                _SmulAddPack_2SW_ACC(*iseq1Ptr++, *iseq2Ptr++);
+                _SmulAddPack_2SW_ACC(*iseq1Ptr++, *iseq2Ptr++);
+                _SmulAddPack_2SW_ACC(*iseq1Ptr++, *iseq2Ptr++);
+                _SmulAddPack_2SW_ACC(*iseq1Ptr++, *iseq2Ptr++);
+            }
+
+        }
+
+        macc40 = _ReadCoProcessor(0);
+        *CrossCorrPtr = (WebRtc_Word32)(macc40 >> right_shifts);
+        CrossCorrPtr += step_seq2;
+    }
+#else // #ifdef _XSCALE_OPT_
+#ifdef _ARM_OPT_
+    WebRtc_Word16 dim_seq8 = (dim_seq >> 3) << 3;
+#endif
+
+    CrossCorrPtr = cross_correlation;
+
+    for (i = 0; i < dim_cross_correlation; i++)
+    {
+        // Set the pointer to the static vector, set the pointer to the sliding vector
+        // and initialize cross_correlation
+        seq1Ptr = seq1;
+        seq2Ptr = seq2 + (step_seq2 * i);
+        (*CrossCorrPtr) = 0;
+
+#ifndef _ARM_OPT_ 
+#ifdef _WIN32
+#pragma message("NOTE: default implementation is used")
+#endif
+        // Perform the cross correlation
+        for (j = 0; j < dim_seq; j++)
+        {
+            (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16_RSFT((*seq1Ptr), (*seq2Ptr), right_shifts);
+            seq1Ptr++;
+            seq2Ptr++;
+        }
+#else
+#ifdef _WIN32
+#pragma message("NOTE: _ARM_OPT_ optimizations are used")
+#endif
+        if (right_shifts == 0)
+        {
+            // Perform the optimized cross correlation
+            for (j = 0; j < dim_seq8; j = j + 8)
+            {
+                (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16((*seq1Ptr), (*seq2Ptr));
+                seq1Ptr++;
+                seq2Ptr++;
+                (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16((*seq1Ptr), (*seq2Ptr));
+                seq1Ptr++;
+                seq2Ptr++;
+                (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16((*seq1Ptr), (*seq2Ptr));
+                seq1Ptr++;
+                seq2Ptr++;
+                (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16((*seq1Ptr), (*seq2Ptr));
+                seq1Ptr++;
+                seq2Ptr++;
+                (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16((*seq1Ptr), (*seq2Ptr));
+                seq1Ptr++;
+                seq2Ptr++;
+                (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16((*seq1Ptr), (*seq2Ptr));
+                seq1Ptr++;
+                seq2Ptr++;
+                (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16((*seq1Ptr), (*seq2Ptr));
+                seq1Ptr++;
+                seq2Ptr++;
+                (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16((*seq1Ptr), (*seq2Ptr));
+                seq1Ptr++;
+                seq2Ptr++;
+            }
+
+            for (j = dim_seq8; j < dim_seq; j++)
+            {
+                (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16((*seq1Ptr), (*seq2Ptr));
+                seq1Ptr++;
+                seq2Ptr++;
+            }
+        }
+        else // right_shifts != 0
+
+        {
+            // Perform the optimized cross correlation
+            for (j = 0; j < dim_seq8; j = j + 8)
+            {
+                (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16_RSFT((*seq1Ptr), (*seq2Ptr),
+                                                             right_shifts);
+                seq1Ptr++;
+                seq2Ptr++;
+                (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16_RSFT((*seq1Ptr), (*seq2Ptr),
+                                                             right_shifts);
+                seq1Ptr++;
+                seq2Ptr++;
+                (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16_RSFT((*seq1Ptr), (*seq2Ptr),
+                                                             right_shifts);
+                seq1Ptr++;
+                seq2Ptr++;
+                (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16_RSFT((*seq1Ptr), (*seq2Ptr),
+                                                             right_shifts);
+                seq1Ptr++;
+                seq2Ptr++;
+                (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16_RSFT((*seq1Ptr), (*seq2Ptr),
+                                                             right_shifts);
+                seq1Ptr++;
+                seq2Ptr++;
+                (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16_RSFT((*seq1Ptr), (*seq2Ptr),
+                                                             right_shifts);
+                seq1Ptr++;
+                seq2Ptr++;
+                (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16_RSFT((*seq1Ptr), (*seq2Ptr),
+                                                             right_shifts);
+                seq1Ptr++;
+                seq2Ptr++;
+                (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16_RSFT((*seq1Ptr), (*seq2Ptr),
+                                                             right_shifts);
+                seq1Ptr++;
+                seq2Ptr++;
+            }
+
+            for (j = dim_seq8; j < dim_seq; j++)
+            {
+                (*CrossCorrPtr) += WEBRTC_SPL_MUL_16_16_RSFT((*seq1Ptr), (*seq2Ptr),
+                                                             right_shifts);
+                seq1Ptr++;
+                seq2Ptr++;
+            }
+        }
+#endif
+        CrossCorrPtr++;
+    }
+#endif
+}

src/common_audio/signal_processing_library/main/source/division_operations.c

@@ -0,0 +1,144 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains implementations of the divisions
+ * WebRtcSpl_DivU32U16()
+ * WebRtcSpl_DivW32W16()
+ * WebRtcSpl_DivW32W16ResW16()
+ * WebRtcSpl_DivResultInQ31()
+ * WebRtcSpl_DivW32HiLow()
+ *
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+WebRtc_UWord32 WebRtcSpl_DivU32U16(WebRtc_UWord32 num, WebRtc_UWord16 den)
+{
+    // Guard against division with 0
+    if (den != 0)
+    {
+        return (WebRtc_UWord32)(num / den);
+    } else
+    {
+        return (WebRtc_UWord32)0xFFFFFFFF;
+    }
+}
+
+WebRtc_Word32 WebRtcSpl_DivW32W16(WebRtc_Word32 num, WebRtc_Word16 den)
+{
+    // Guard against division with 0
+    if (den != 0)
+    {
+        return (WebRtc_Word32)(num / den);
+    } else
+    {
+        return (WebRtc_Word32)0x7FFFFFFF;
+    }
+}
+
+WebRtc_Word16 WebRtcSpl_DivW32W16ResW16(WebRtc_Word32 num, WebRtc_Word16 den)
+{
+    // Guard against division with 0
+    if (den != 0)
+    {
+        return (WebRtc_Word16)(num / den);
+    } else
+    {
+        return (WebRtc_Word16)0x7FFF;
+    }
+}
+
+WebRtc_Word32 WebRtcSpl_DivResultInQ31(WebRtc_Word32 num, WebRtc_Word32 den)
+{
+    WebRtc_Word32 L_num = num;
+    WebRtc_Word32 L_den = den;
+    WebRtc_Word32 div = 0;
+    int k = 31;
+    int change_sign = 0;
+
+    if (num == 0)
+        return 0;
+
+    if (num < 0)
+    {
+        change_sign++;
+        L_num = -num;
+    }
+    if (den < 0)
+    {
+        change_sign++;
+        L_den = -den;
+    }
+    while (k--)
+    {
+        div <<= 1;
+        L_num <<= 1;
+        if (L_num >= L_den)
+        {
+            L_num -= L_den;
+            div++;
+        }
+    }
+    if (change_sign == 1)
+    {
+        div = -div;
+    }
+    return div;
+}
+
+WebRtc_Word32 WebRtcSpl_DivW32HiLow(WebRtc_Word32 num, WebRtc_Word16 den_hi,
+                                    WebRtc_Word16 den_low)
+{
+    WebRtc_Word16 approx, tmp_hi, tmp_low, num_hi, num_low;
+    WebRtc_Word32 tmpW32;
+
+    approx = (WebRtc_Word16)WebRtcSpl_DivW32W16((WebRtc_Word32)0x1FFFFFFF, den_hi);
+    // result in Q14 (Note: 3FFFFFFF = 0.5 in Q30)
+
+    // tmpW32 = 1/den = approx * (2.0 - den * approx) (in Q30)
+    tmpW32 = (WEBRTC_SPL_MUL_16_16(den_hi, approx) << 1)
+            + ((WEBRTC_SPL_MUL_16_16(den_low, approx) >> 15) << 1);
+    // tmpW32 = den * approx
+
+    tmpW32 = (WebRtc_Word32)0x7fffffffL - tmpW32; // result in Q30 (tmpW32 = 2.0-(den*approx))
+
+    // Store tmpW32 in hi and low format
+    tmp_hi = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmpW32, 16);
+    tmp_low = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32((tmpW32
+            - WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)tmp_hi, 16)), 1);
+
+    // tmpW32 = 1/den in Q29
+    tmpW32 = ((WEBRTC_SPL_MUL_16_16(tmp_hi, approx) + (WEBRTC_SPL_MUL_16_16(tmp_low, approx)
+            >> 15)) << 1);
+
+    // 1/den in hi and low format
+    tmp_hi = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmpW32, 16);
+    tmp_low = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32((tmpW32
+            - WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)tmp_hi, 16)), 1);
+
+    // Store num in hi and low format
+    num_hi = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(num, 16);
+    num_low = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32((num
+            - WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)num_hi, 16)), 1);
+
+    // num * (1/den) by 32 bit multiplication (result in Q28)
+
+    tmpW32 = (WEBRTC_SPL_MUL_16_16(num_hi, tmp_hi) + (WEBRTC_SPL_MUL_16_16(num_hi, tmp_low)
+            >> 15) + (WEBRTC_SPL_MUL_16_16(num_low, tmp_hi) >> 15));
+
+    // Put result in Q31 (convert from Q28)
+    tmpW32 = WEBRTC_SPL_LSHIFT_W32(tmpW32, 3);
+
+    return tmpW32;
+}

src/common_audio/signal_processing_library/main/source/dot_product_with_scale.c

@@ -0,0 +1,91 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_DotProductWithScale().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+WebRtc_Word32 WebRtcSpl_DotProductWithScale(WebRtc_Word16 *vector1, WebRtc_Word16 *vector2,
+                                            int length, int scaling)
+{
+    WebRtc_Word32 sum;
+    int i;
+#ifdef _ARM_OPT_
+#pragma message("NOTE: _ARM_OPT_ optimizations are used")
+    WebRtc_Word16 len4 = (length >> 2) << 2;
+#endif
+
+    sum = 0;
+
+#ifndef _ARM_OPT_
+    for (i = 0; i < length; i++)
+    {
+        sum += WEBRTC_SPL_MUL_16_16_RSFT(*vector1++, *vector2++, scaling);
+    }
+#else
+    if (scaling == 0)
+    {
+        for (i = 0; i < len4; i = i + 4)
+        {
+            sum += WEBRTC_SPL_MUL_16_16(*vector1, *vector2);
+            vector1++;
+            vector2++;
+            sum += WEBRTC_SPL_MUL_16_16(*vector1, *vector2);
+            vector1++;
+            vector2++;
+            sum += WEBRTC_SPL_MUL_16_16(*vector1, *vector2);
+            vector1++;
+            vector2++;
+            sum += WEBRTC_SPL_MUL_16_16(*vector1, *vector2);
+            vector1++;
+            vector2++;
+        }
+
+        for (i = len4; i < length; i++)
+        {
+            sum += WEBRTC_SPL_MUL_16_16(*vector1, *vector2);
+            vector1++;
+            vector2++;
+        }
+    }
+    else
+    {
+        for (i = 0; i < len4; i = i + 4)
+        {
+            sum += WEBRTC_SPL_MUL_16_16_RSFT(*vector1, *vector2, scaling);
+            vector1++;
+            vector2++;
+            sum += WEBRTC_SPL_MUL_16_16_RSFT(*vector1, *vector2, scaling);
+            vector1++;
+            vector2++;
+            sum += WEBRTC_SPL_MUL_16_16_RSFT(*vector1, *vector2, scaling);
+            vector1++;
+            vector2++;
+            sum += WEBRTC_SPL_MUL_16_16_RSFT(*vector1, *vector2, scaling);
+            vector1++;
+            vector2++;
+        }
+
+        for (i = len4; i < length; i++)
+        {
+            sum += WEBRTC_SPL_MUL_16_16_RSFT(*vector1, *vector2, scaling);
+            vector1++;
+            vector2++;
+        }
+    }
+#endif
+
+    return sum;
+}

src/common_audio/signal_processing_library/main/source/downsample_fast.c

@@ -0,0 +1,59 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_DownsampleFast().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+int WebRtcSpl_DownsampleFast(WebRtc_Word16 *in_ptr, WebRtc_Word16 in_length,
+                             WebRtc_Word16 *out_ptr, WebRtc_Word16 out_length,
+                             WebRtc_Word16 *B, WebRtc_Word16 B_length, WebRtc_Word16 factor,
+                             WebRtc_Word16 delay)
+{
+    WebRtc_Word32 o;
+    int i, j;
+
+    WebRtc_Word16 *downsampled_ptr = out_ptr;
+    WebRtc_Word16 *b_ptr;
+    WebRtc_Word16 *x_ptr;
+    WebRtc_Word16 endpos = delay
+            + (WebRtc_Word16)WEBRTC_SPL_MUL_16_16(factor, (out_length - 1)) + 1;
+
+    if (in_length < endpos)
+    {
+        return -1;
+    }
+
+    for (i = delay; i < endpos; i += factor)
+    {
+        b_ptr = &B[0];
+        x_ptr = &in_ptr[i];
+
+        o = (WebRtc_Word32)2048; // Round val
+
+        for (j = 0; j < B_length; j++)
+        {
+            o += WEBRTC_SPL_MUL_16_16(*b_ptr++, *x_ptr--);
+        }
+
+        o = WEBRTC_SPL_RSHIFT_W32(o, 12);
+
+        // If output is higher than 32768, saturate it. Same with negative side
+
+        *downsampled_ptr++ = (WebRtc_Word16)WEBRTC_SPL_SAT(32767, o, -32768);
+    }
+
+    return 0;
+}

src/common_audio/signal_processing_library/main/source/energy.c

@@ -0,0 +1,36 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_Energy().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+WebRtc_Word32 WebRtcSpl_Energy(WebRtc_Word16* vector, int vector_length, int* scale_factor)
+{
+    WebRtc_Word32 en = 0;
+    int i;
+    int scaling = WebRtcSpl_GetScalingSquare(vector, vector_length, vector_length);
+    int looptimes = vector_length;
+    WebRtc_Word16 *vectorptr = vector;
+
+    for (i = 0; i < looptimes; i++)
+    {
+        en += WEBRTC_SPL_MUL_16_16_RSFT(*vectorptr, *vectorptr, scaling);
+        vectorptr++;
+    }
+    *scale_factor = scaling;
+
+    return en;
+}

src/common_audio/signal_processing_library/main/source/filter_ar.c

@@ -0,0 +1,89 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_FilterAR().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+int WebRtcSpl_FilterAR(G_CONST WebRtc_Word16* a,
+                       int a_length,
+                       G_CONST WebRtc_Word16* x,
+                       int x_length,
+                       WebRtc_Word16* state,
+                       int state_length,
+                       WebRtc_Word16* state_low,
+                       int state_low_length,
+                       WebRtc_Word16* filtered,
+                       WebRtc_Word16* filtered_low,
+                       int filtered_low_length)
+{
+    WebRtc_Word32 o;
+    WebRtc_Word32 oLOW;
+    int i, j, stop;
+    G_CONST WebRtc_Word16* x_ptr = &x[0];
+    WebRtc_Word16* filteredFINAL_ptr = filtered;
+    WebRtc_Word16* filteredFINAL_LOW_ptr = filtered_low;
+
+    for (i = 0; i < x_length; i++)
+    {
+        // Calculate filtered[i] and filtered_low[i]
+        G_CONST WebRtc_Word16* a_ptr = &a[1];
+        WebRtc_Word16* filtered_ptr = &filtered[i - 1];
+        WebRtc_Word16* filtered_low_ptr = &filtered_low[i - 1];
+        WebRtc_Word16* state_ptr = &state[state_length - 1];
+        WebRtc_Word16* state_low_ptr = &state_low[state_length - 1];
+
+        o = (WebRtc_Word32)(*x_ptr++) << 12;
+        oLOW = (WebRtc_Word32)0;
+
+        stop = (i < a_length) ? i + 1 : a_length;
+        for (j = 1; j < stop; j++)
+        {
+            o -= WEBRTC_SPL_MUL_16_16(*a_ptr, *filtered_ptr--);
+            oLOW -= WEBRTC_SPL_MUL_16_16(*a_ptr++, *filtered_low_ptr--);
+        }
+        for (j = i + 1; j < a_length; j++)
+        {
+            o -= WEBRTC_SPL_MUL_16_16(*a_ptr, *state_ptr--);
+            oLOW -= WEBRTC_SPL_MUL_16_16(*a_ptr++, *state_low_ptr--);
+        }
+
+        o += (oLOW >> 12);
+        *filteredFINAL_ptr = (WebRtc_Word16)((o + (WebRtc_Word32)2048) >> 12);
+        *filteredFINAL_LOW_ptr++ = (WebRtc_Word16)(o - ((WebRtc_Word32)(*filteredFINAL_ptr++)
+                << 12));
+    }
+
+    // Save the filter state
+    if (x_length >= state_length)
+    {
+        WebRtcSpl_CopyFromEndW16(filtered, x_length, a_length - 1, state);
+        WebRtcSpl_CopyFromEndW16(filtered_low, x_length, a_length - 1, state_low);
+    } else
+    {
+        for (i = 0; i < state_length - x_length; i++)
+        {
+            state[i] = state[i + x_length];
+            state_low[i] = state_low[i + x_length];
+        }
+        for (i = 0; i < x_length; i++)
+        {
+            state[state_length - x_length + i] = filtered[i];
+            state[state_length - x_length + i] = filtered_low[i];
+        }
+    }
+
+    return x_length;
+}

src/common_audio/signal_processing_library/main/source/filter_ar_fast_q12.c

@@ -0,0 +1,49 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_FilterARFastQ12().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+void WebRtcSpl_FilterARFastQ12(WebRtc_Word16 *in, WebRtc_Word16 *out, WebRtc_Word16 *A,
+                               WebRtc_Word16 A_length, WebRtc_Word16 length)
+{
+    WebRtc_Word32 o;
+    int i, j;
+
+    WebRtc_Word16 *x_ptr = &in[0];
+    WebRtc_Word16 *filtered_ptr = &out[0];
+
+    for (i = 0; i < length; i++)
+    {
+        // Calculate filtered[i]
+        G_CONST WebRtc_Word16 *a_ptr = &A[0];
+        WebRtc_Word16 *state_ptr = &out[i - 1];
+
+        o = WEBRTC_SPL_MUL_16_16(*x_ptr++, *a_ptr++);
+
+        for (j = 1; j < A_length; j++)
+        {
+            o -= WEBRTC_SPL_MUL_16_16(*a_ptr++,*state_ptr--);
+        }
+
+        // Saturate the output
+        o = WEBRTC_SPL_SAT((WebRtc_Word32)134215679, o, (WebRtc_Word32)-134217728);
+
+        *filtered_ptr++ = (WebRtc_Word16)((o + (WebRtc_Word32)2048) >> 12);
+    }
+
+    return;
+}

src/common_audio/signal_processing_library/main/source/filter_ma_fast_q12.c

@@ -0,0 +1,49 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_FilterMAFastQ12().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+void WebRtcSpl_FilterMAFastQ12(WebRtc_Word16* in_ptr,
+                               WebRtc_Word16* out_ptr,
+                               WebRtc_Word16* B,
+                               WebRtc_Word16 B_length,
+                               WebRtc_Word16 length)
+{
+    WebRtc_Word32 o;
+    int i, j;
+    for (i = 0; i < length; i++)
+    {
+        G_CONST WebRtc_Word16* b_ptr = &B[0];
+        G_CONST WebRtc_Word16* x_ptr = &in_ptr[i];
+
+        o = (WebRtc_Word32)0;
+
+        for (j = 0; j < B_length; j++)
+        {
+            o += WEBRTC_SPL_MUL_16_16(*b_ptr++, *x_ptr--);
+        }
+
+        // If output is higher than 32768, saturate it. Same with negative side
+        // 2^27 = 134217728, which corresponds to 32768 in Q12
+
+        // Saturate the output
+        o = WEBRTC_SPL_SAT((WebRtc_Word32)134215679, o, (WebRtc_Word32)-134217728);
+
+        *out_ptr++ = (WebRtc_Word16)((o + (WebRtc_Word32)2048) >> 12);
+    }
+    return;
+}

src/common_audio/signal_processing_library/main/source/get_hanning_window.c

@@ -0,0 +1,41 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_GetHanningWindow().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+void WebRtcSpl_GetHanningWindow(WebRtc_Word16 *v, WebRtc_Word16 size)
+{
+    int jj;
+    WebRtc_Word16 *vptr1;
+
+    WebRtc_Word32 index;
+    WebRtc_Word32 factor = ((WebRtc_Word32)0x40000000);
+
+    factor = WebRtcSpl_DivW32W16(factor, size);
+    if (size < 513)
+        index = (WebRtc_Word32)-0x200000;
+    else
+        index = (WebRtc_Word32)-0x100000;
+    vptr1 = v;
+
+    for (jj = 0; jj < size; jj++)
+    {
+        index += factor;
+        (*vptr1++) = WebRtcSpl_kHanningTable[index >> 22];
+    }
+
+}

src/common_audio/signal_processing_library/main/source/get_scaling_square.c

@@ -0,0 +1,44 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_GetScalingSquare().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+int WebRtcSpl_GetScalingSquare(WebRtc_Word16 *in_vector, int in_vector_length, int times)
+{
+    int nbits = WebRtcSpl_GetSizeInBits(times);
+    int i;
+    WebRtc_Word16 smax = -1;
+    WebRtc_Word16 sabs;
+    WebRtc_Word16 *sptr = in_vector;
+    int t;
+    int looptimes = in_vector_length;
+
+    for (i = looptimes; i > 0; i--)
+    {
+        sabs = (*sptr > 0 ? *sptr++ : -*sptr++);
+        smax = (sabs > smax ? sabs : smax);
+    }
+    t = WebRtcSpl_NormW32(WEBRTC_SPL_MUL(smax, smax));
+
+    if (smax == 0)
+    {
+        return 0; // Since norm(0) returns 0
+    } else
+    {
+        return (t > nbits) ? 0 : nbits - t;
+    }
+}

src/common_audio/signal_processing_library/main/source/hanning_table.c

@@ -0,0 +1,53 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the Hanning table with 256 entries.
+ *
+ */
+
+#include "signal_processing_library.h"
+
+// Hanning table with 256 entries
+WebRtc_Word16 WebRtcSpl_kHanningTable[] = {
+    1,      2,      6,     10,     15,     22,     30,     39,
+   50,     62,     75,     89,    104,    121,    138,    157,
+  178,    199,    222,    246,    271,    297,    324,    353,
+  383,    413,    446,    479,    513,    549,    586,    624,
+  663,    703,    744,    787,    830,    875,    920,    967,
+ 1015,   1064,   1114,   1165,   1218,   1271,   1325,   1381,
+ 1437,   1494,   1553,   1612,   1673,   1734,   1796,   1859,
+ 1924,   1989,   2055,   2122,   2190,   2259,   2329,   2399,
+ 2471,   2543,   2617,   2691,   2765,   2841,   2918,   2995,
+ 3073,   3152,   3232,   3312,   3393,   3475,   3558,   3641,
+ 3725,   3809,   3895,   3980,   4067,   4154,   4242,   4330,
+ 4419,   4509,   4599,   4689,   4781,   4872,   4964,   5057,
+ 5150,   5244,   5338,   5432,   5527,   5622,   5718,   5814,
+ 5910,   6007,   6104,   6202,   6299,   6397,   6495,   6594,
+ 6693,   6791,   6891,   6990,   7090,   7189,   7289,   7389,
+ 7489,   7589,   7690,   7790,   7890,   7991,   8091,   8192,
+ 8293,   8393,   8494,   8594,   8694,   8795,   8895,   8995,
+ 9095,   9195,   9294,   9394,   9493,   9593,   9691,   9790,
+ 9889,   9987,  10085,  10182,  10280,  10377,  10474,  10570,
+10666,  10762,  10857,  10952,  11046,  11140,  11234,  11327,
+11420,  11512,  11603,  11695,  11785,  11875,  11965,  12054,
+12142,  12230,  12317,  12404,  12489,  12575,  12659,  12743,
+12826,  12909,  12991,  13072,  13152,  13232,  13311,  13389,
+13466,  13543,  13619,  13693,  13767,  13841,  13913,  13985,
+14055,  14125,  14194,  14262,  14329,  14395,  14460,  14525,
+14588,  14650,  14711,  14772,  14831,  14890,  14947,  15003,
+15059,  15113,  15166,  15219,  15270,  15320,  15369,  15417,
+15464,  15509,  15554,  15597,  15640,  15681,  15721,  15760,
+15798,  15835,  15871,  15905,  15938,  15971,  16001,  16031,
+16060,  16087,  16113,  16138,  16162,  16185,  16206,  16227,
+16246,  16263,  16280,  16295,  16309,  16322,  16334,  16345,
+16354,  16362,  16369,  16374,  16378,  16382,  16383,  16384
+};

src/common_audio/signal_processing_library/main/source/ilbc_specific_functions.c

@@ -0,0 +1,120 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains implementations of the iLBC specific functions
+ * WebRtcSpl_ScaleAndAddVectorsWithRound()
+ * WebRtcSpl_ReverseOrderMultArrayElements()
+ * WebRtcSpl_ElementwiseVectorMult()
+ * WebRtcSpl_AddVectorsAndShift()
+ * WebRtcSpl_AddAffineVectorToVector()
+ * WebRtcSpl_AffineTransformVector()
+ *
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+void WebRtcSpl_ScaleAndAddVectorsWithRound(WebRtc_Word16 *vector1, WebRtc_Word16 scale1,
+                                           WebRtc_Word16 *vector2, WebRtc_Word16 scale2,
+                                           WebRtc_Word16 right_shifts, WebRtc_Word16 *out,
+                                           WebRtc_Word16 vector_length)
+{
+    int i;
+    WebRtc_Word16 roundVal;
+    roundVal = 1 << right_shifts;
+    roundVal = roundVal >> 1;
+    for (i = 0; i < vector_length; i++)
+    {
+        out[i] = (WebRtc_Word16)((WEBRTC_SPL_MUL_16_16(vector1[i], scale1)
+                + WEBRTC_SPL_MUL_16_16(vector2[i], scale2) + roundVal) >> right_shifts);
+    }
+}
+
+void WebRtcSpl_ReverseOrderMultArrayElements(WebRtc_Word16 *out, G_CONST WebRtc_Word16 *in,
+                                             G_CONST WebRtc_Word16 *win,
+                                             WebRtc_Word16 vector_length,
+                                             WebRtc_Word16 right_shifts)
+{
+    int i;
+    WebRtc_Word16 *outptr = out;
+    G_CONST WebRtc_Word16 *inptr = in;
+    G_CONST WebRtc_Word16 *winptr = win;
+    for (i = 0; i < vector_length; i++)
+    {
+        (*outptr++) = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(*inptr++,
+                                                               *winptr--, right_shifts);
+    }
+}
+
+void WebRtcSpl_ElementwiseVectorMult(WebRtc_Word16 *out, G_CONST WebRtc_Word16 *in,
+                                     G_CONST WebRtc_Word16 *win, WebRtc_Word16 vector_length,
+                                     WebRtc_Word16 right_shifts)
+{
+    int i;
+    WebRtc_Word16 *outptr = out;
+    G_CONST WebRtc_Word16 *inptr = in;
+    G_CONST WebRtc_Word16 *winptr = win;
+    for (i = 0; i < vector_length; i++)
+    {
+        (*outptr++) = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(*inptr++,
+                                                               *winptr++, right_shifts);
+    }
+}
+
+void WebRtcSpl_AddVectorsAndShift(WebRtc_Word16 *out, G_CONST WebRtc_Word16 *in1,
+                                  G_CONST WebRtc_Word16 *in2, WebRtc_Word16 vector_length,
+                                  WebRtc_Word16 right_shifts)
+{
+    int i;
+    WebRtc_Word16 *outptr = out;
+    G_CONST WebRtc_Word16 *in1ptr = in1;
+    G_CONST WebRtc_Word16 *in2ptr = in2;
+    for (i = vector_length; i > 0; i--)
+    {
+        (*outptr++) = (WebRtc_Word16)(((*in1ptr++) + (*in2ptr++)) >> right_shifts);
+    }
+}
+
+void WebRtcSpl_AddAffineVectorToVector(WebRtc_Word16 *out, WebRtc_Word16 *in,
+                                       WebRtc_Word16 gain, WebRtc_Word32 add_constant,
+                                       WebRtc_Word16 right_shifts, int vector_length)
+{
+    WebRtc_Word16 *inPtr;
+    WebRtc_Word16 *outPtr;
+    int i;
+
+    inPtr = in;
+    outPtr = out;
+    for (i = 0; i < vector_length; i++)
+    {
+        (*outPtr++) += (WebRtc_Word16)((WEBRTC_SPL_MUL_16_16((*inPtr++), gain)
+                + (WebRtc_Word32)add_constant) >> right_shifts);
+    }
+}
+
+void WebRtcSpl_AffineTransformVector(WebRtc_Word16 *out, WebRtc_Word16 *in,
+                                     WebRtc_Word16 gain, WebRtc_Word32 add_constant,
+                                     WebRtc_Word16 right_shifts, int vector_length)
+{
+    WebRtc_Word16 *inPtr;
+    WebRtc_Word16 *outPtr;
+    int i;
+
+    inPtr = in;
+    outPtr = out;
+    for (i = 0; i < vector_length; i++)
+    {
+        (*outPtr++) = (WebRtc_Word16)((WEBRTC_SPL_MUL_16_16((*inPtr++), gain)
+                + (WebRtc_Word32)add_constant) >> right_shifts);
+    }
+}

src/common_audio/signal_processing_library/main/source/levinson_durbin.c

@@ -0,0 +1,259 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_LevinsonDurbin().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+#define SPL_LEVINSON_MAXORDER 20
+
+WebRtc_Word16 WebRtcSpl_LevinsonDurbin(WebRtc_Word32 *R, WebRtc_Word16 *A, WebRtc_Word16 *K,
+                                       WebRtc_Word16 order)
+{
+    WebRtc_Word16 i, j;
+    // Auto-correlation coefficients in high precision
+    WebRtc_Word16 R_hi[SPL_LEVINSON_MAXORDER + 1], R_low[SPL_LEVINSON_MAXORDER + 1];
+    // LPC coefficients in high precision
+    WebRtc_Word16 A_hi[SPL_LEVINSON_MAXORDER + 1], A_low[SPL_LEVINSON_MAXORDER + 1];
+    // LPC coefficients for next iteration
+    WebRtc_Word16 A_upd_hi[SPL_LEVINSON_MAXORDER + 1], A_upd_low[SPL_LEVINSON_MAXORDER + 1];
+    // Reflection coefficient in high precision
+    WebRtc_Word16 K_hi, K_low;
+    // Prediction gain Alpha in high precision and with scale factor
+    WebRtc_Word16 Alpha_hi, Alpha_low, Alpha_exp;
+    WebRtc_Word16 tmp_hi, tmp_low;
+    WebRtc_Word32 temp1W32, temp2W32, temp3W32;
+    WebRtc_Word16 norm;
+
+    // Normalize the autocorrelation R[0]...R[order+1]
+
+    norm = WebRtcSpl_NormW32(R[0]);
+
+    for (i = order; i >= 0; i--)
+    {
+        temp1W32 = WEBRTC_SPL_LSHIFT_W32(R[i], norm);
+        // Put R in hi and low format
+        R_hi[i] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(temp1W32, 16);
+        R_low[i] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32((temp1W32
+                - WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)R_hi[i], 16)), 1);
+    }
+
+    // K = A[1] = -R[1] / R[0]
+
+    temp2W32 = WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)R_hi[1],16)
+            + WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)R_low[1],1); // R[1] in Q31
+    temp3W32 = WEBRTC_SPL_ABS_W32(temp2W32); // abs R[1]
+    temp1W32 = WebRtcSpl_DivW32HiLow(temp3W32, R_hi[0], R_low[0]); // abs(R[1])/R[0] in Q31
+    // Put back the sign on R[1]
+    if (temp2W32 > 0)
+    {
+        temp1W32 = -temp1W32;
+    }
+
+    // Put K in hi and low format
+    K_hi = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(temp1W32, 16);
+    K_low = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32((temp1W32
+            - WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)K_hi, 16)), 1);
+
+    // Store first reflection coefficient
+    K[0] = K_hi;
+
+    temp1W32 = WEBRTC_SPL_RSHIFT_W32(temp1W32, 4); // A[1] in Q27
+
+    // Put A[1] in hi and low format
+    A_hi[1] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(temp1W32, 16);
+    A_low[1] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32((temp1W32
+            - WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)A_hi[1], 16)), 1);
+
+    // Alpha = R[0] * (1-K^2)
+
+    temp1W32 = (((WEBRTC_SPL_MUL_16_16(K_hi, K_low) >> 14) + WEBRTC_SPL_MUL_16_16(K_hi, K_hi))
+            << 1); // temp1W32 = k^2 in Q31
+
+    temp1W32 = WEBRTC_SPL_ABS_W32(temp1W32); // Guard against <0
+    temp1W32 = (WebRtc_Word32)0x7fffffffL - temp1W32; // temp1W32 = (1 - K[0]*K[0]) in Q31
+
+    // Store temp1W32 = 1 - K[0]*K[0] on hi and low format
+    tmp_hi = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(temp1W32, 16);
+    tmp_low = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32((temp1W32
+            - WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)tmp_hi, 16)), 1);
+
+    // Calculate Alpha in Q31
+    temp1W32 = ((WEBRTC_SPL_MUL_16_16(R_hi[0], tmp_hi)
+            + (WEBRTC_SPL_MUL_16_16(R_hi[0], tmp_low) >> 15)
+            + (WEBRTC_SPL_MUL_16_16(R_low[0], tmp_hi) >> 15)) << 1);
+
+    // Normalize Alpha and put it in hi and low format
+
+    Alpha_exp = WebRtcSpl_NormW32(temp1W32);
+    temp1W32 = WEBRTC_SPL_LSHIFT_W32(temp1W32, Alpha_exp);
+    Alpha_hi = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(temp1W32, 16);
+    Alpha_low = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32((temp1W32
+            - WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)Alpha_hi, 16)), 1);
+
+    // Perform the iterative calculations in the Levinson-Durbin algorithm
+
+    for (i = 2; i <= order; i++)
+    {
+        /*                    ----
+         temp1W32 =  R[i] + > R[j]*A[i-j]
+         /
+         ----
+         j=1..i-1
+         */
+
+        temp1W32 = 0;
+
+        for (j = 1; j < i; j++)
+        {
+            // temp1W32 is in Q31
+            temp1W32 += ((WEBRTC_SPL_MUL_16_16(R_hi[j], A_hi[i-j]) << 1)
+                    + (((WEBRTC_SPL_MUL_16_16(R_hi[j], A_low[i-j]) >> 15)
+                            + (WEBRTC_SPL_MUL_16_16(R_low[j], A_hi[i-j]) >> 15)) << 1));
+        }
+
+        temp1W32 = WEBRTC_SPL_LSHIFT_W32(temp1W32, 4);
+        temp1W32 += (WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)R_hi[i], 16)
+                + WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)R_low[i], 1));
+
+        // K = -temp1W32 / Alpha
+        temp2W32 = WEBRTC_SPL_ABS_W32(temp1W32); // abs(temp1W32)
+        temp3W32 = WebRtcSpl_DivW32HiLow(temp2W32, Alpha_hi, Alpha_low); // abs(temp1W32)/Alpha
+
+        // Put the sign of temp1W32 back again
+        if (temp1W32 > 0)
+        {
+            temp3W32 = -temp3W32;
+        }
+
+        // Use the Alpha shifts from earlier to de-normalize
+        norm = WebRtcSpl_NormW32(temp3W32);
+        if ((Alpha_exp <= norm) || (temp3W32 == 0))
+        {
+            temp3W32 = WEBRTC_SPL_LSHIFT_W32(temp3W32, Alpha_exp);
+        } else
+        {
+            if (temp3W32 > 0)
+            {
+                temp3W32 = (WebRtc_Word32)0x7fffffffL;
+            } else
+            {
+                temp3W32 = (WebRtc_Word32)0x80000000L;
+            }
+        }
+
+        // Put K on hi and low format
+        K_hi = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(temp3W32, 16);
+        K_low = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32((temp3W32
+                - WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)K_hi, 16)), 1);
+
+        // Store Reflection coefficient in Q15
+        K[i - 1] = K_hi;
+
+        // Test for unstable filter.
+        // If unstable return 0 and let the user decide what to do in that case
+
+        if ((WebRtc_Word32)WEBRTC_SPL_ABS_W16(K_hi) > (WebRtc_Word32)32750)
+        {
+            return 0; // Unstable filter
+        }
+
+        /*
+         Compute updated LPC coefficient: Anew[i]
+         Anew[j]= A[j] + K*A[i-j]   for j=1..i-1
+         Anew[i]= K
+         */
+
+        for (j = 1; j < i; j++)
+        {
+            // temp1W32 = A[j] in Q27
+            temp1W32 = WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)A_hi[j],16)
+                    + WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)A_low[j],1);
+
+            // temp1W32 += K*A[i-j] in Q27
+            temp1W32 += ((WEBRTC_SPL_MUL_16_16(K_hi, A_hi[i-j])
+                    + (WEBRTC_SPL_MUL_16_16(K_hi, A_low[i-j]) >> 15)
+                    + (WEBRTC_SPL_MUL_16_16(K_low, A_hi[i-j]) >> 15)) << 1);
+
+            // Put Anew in hi and low format
+            A_upd_hi[j] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(temp1W32, 16);
+            A_upd_low[j] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32((temp1W32
+                    - WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)A_upd_hi[j], 16)), 1);
+        }
+
+        // temp3W32 = K in Q27 (Convert from Q31 to Q27)
+        temp3W32 = WEBRTC_SPL_RSHIFT_W32(temp3W32, 4);
+
+        // Store Anew in hi and low format
+        A_upd_hi[i] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(temp3W32, 16);
+        A_upd_low[i] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32((temp3W32
+                - WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)A_upd_hi[i], 16)), 1);
+
+        // Alpha = Alpha * (1-K^2)
+
+        temp1W32 = (((WEBRTC_SPL_MUL_16_16(K_hi, K_low) >> 14)
+                + WEBRTC_SPL_MUL_16_16(K_hi, K_hi)) << 1); // K*K in Q31
+
+        temp1W32 = WEBRTC_SPL_ABS_W32(temp1W32); // Guard against <0
+        temp1W32 = (WebRtc_Word32)0x7fffffffL - temp1W32; // 1 - K*K  in Q31
+
+        // Convert 1- K^2 in hi and low format
+        tmp_hi = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(temp1W32, 16);
+        tmp_low = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32((temp1W32
+                - WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)tmp_hi, 16)), 1);
+
+        // Calculate Alpha = Alpha * (1-K^2) in Q31
+        temp1W32 = ((WEBRTC_SPL_MUL_16_16(Alpha_hi, tmp_hi)
+                + (WEBRTC_SPL_MUL_16_16(Alpha_hi, tmp_low) >> 15)
+                + (WEBRTC_SPL_MUL_16_16(Alpha_low, tmp_hi) >> 15)) << 1);
+
+        // Normalize Alpha and store it on hi and low format
+
+        norm = WebRtcSpl_NormW32(temp1W32);
+        temp1W32 = WEBRTC_SPL_LSHIFT_W32(temp1W32, norm);
+
+        Alpha_hi = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(temp1W32, 16);
+        Alpha_low = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32((temp1W32
+                - WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)Alpha_hi, 16)), 1);
+
+        // Update the total normalization of Alpha
+        Alpha_exp = Alpha_exp + norm;
+
+        // Update A[]
+
+        for (j = 1; j <= i; j++)
+        {
+            A_hi[j] = A_upd_hi[j];
+            A_low[j] = A_upd_low[j];
+        }
+    }
+
+    /*
+     Set A[0] to 1.0 and store the A[i] i=1...order in Q12
+     (Convert from Q27 and use rounding)
+     */
+
+    A[0] = 4096;
+
+    for (i = 1; i <= order; i++)
+    {
+        // temp1W32 in Q27
+        temp1W32 = WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)A_hi[i], 16)
+                + WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)A_low[i], 1);
+        // Round and store upper word
+        A[i] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32((temp1W32<<1)+(WebRtc_Word32)32768, 16);
+    }
+    return 1; // Stable filters
+}

src/common_audio/signal_processing_library/main/source/lpc_to_refl_coef.c

@@ -0,0 +1,57 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_LpcToReflCoef().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+#define SPL_LPC_TO_REFL_COEF_MAX_AR_MODEL_ORDER 50
+
+void WebRtcSpl_LpcToReflCoef(WebRtc_Word16* a16, int use_order, WebRtc_Word16* k16)
+{
+    int m, k;
+    WebRtc_Word32 tmp32[SPL_LPC_TO_REFL_COEF_MAX_AR_MODEL_ORDER];
+    WebRtc_Word32 tmp_inv_denom32;
+    WebRtc_Word16 tmp_inv_denom16;
+
+    k16[use_order - 1] = WEBRTC_SPL_LSHIFT_W16(a16[use_order], 3); //Q12<<3 => Q15
+    for (m = use_order - 1; m > 0; m--)
+    {
+        // (1 - k^2) in Q30
+        tmp_inv_denom32 = ((WebRtc_Word32)1073741823) - WEBRTC_SPL_MUL_16_16(k16[m], k16[m]);
+        // (1 - k^2) in Q15
+        tmp_inv_denom16 = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp_inv_denom32, 15);
+
+        for (k = 1; k <= m; k++)
+        {
+            // tmp[k] = (a[k] - RC[m] * a[m-k+1]) / (1.0 - RC[m]*RC[m]);
+
+            // [Q12<<16 - (Q15*Q12)<<1] = [Q28 - Q28] = Q28
+            tmp32[k] = WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)a16[k], 16)
+                    - WEBRTC_SPL_LSHIFT_W32(WEBRTC_SPL_MUL_16_16(k16[m], a16[m-k+1]), 1);
+
+            tmp32[k] = WebRtcSpl_DivW32W16(tmp32[k], tmp_inv_denom16); //Q28/Q15 = Q13
+        }
+
+        for (k = 1; k < m; k++)
+        {
+            a16[k] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp32[k], 1); //Q13>>1 => Q12
+        }
+
+        tmp32[m] = WEBRTC_SPL_SAT(8191, tmp32[m], -8191);
+        k16[m - 1] = (WebRtc_Word16)WEBRTC_SPL_LSHIFT_W32(tmp32[m], 2); //Q13<<2 => Q15
+    }
+    return;
+}

src/common_audio/signal_processing_library/main/source/min_max_operations.c

@@ -0,0 +1,265 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+/*
+ * This file contains the implementation of functions
+ * WebRtcSpl_MaxAbsValueW16()
+ * WebRtcSpl_MaxAbsIndexW16()
+ * WebRtcSpl_MaxAbsValueW32()
+ * WebRtcSpl_MaxValueW16()
+ * WebRtcSpl_MaxIndexW16()
+ * WebRtcSpl_MaxValueW32()
+ * WebRtcSpl_MaxIndexW32()
+ * WebRtcSpl_MinValueW16()
+ * WebRtcSpl_MinIndexW16()
+ * WebRtcSpl_MinValueW32()
+ * WebRtcSpl_MinIndexW32()
+ *
+ * The description header can be found in signal_processing_library.h.
+ *
+ */
+
+#include "signal_processing_library.h"
+
+#if !(defined(WEBRTC_ANDROID) && defined(WEBRTC_ARCH_ARM_NEON))
+
+// Maximum absolute value of word16 vector.
+WebRtc_Word16 WebRtcSpl_MaxAbsValueW16(const WebRtc_Word16 *vector, WebRtc_Word16 length)
+{
+    WebRtc_Word32 tempMax = 0;
+    WebRtc_Word32 absVal;
+    WebRtc_Word16 totMax;
+    int i;
+    G_CONST WebRtc_Word16 *tmpvector = vector;
+
+    for (i = 0; i < length; i++)
+    {
+        absVal = WEBRTC_SPL_ABS_W32((*tmpvector));
+        if (absVal > tempMax)
+        {
+            tempMax = absVal;
+        }
+        tmpvector++;
+    }
+    totMax = (WebRtc_Word16)WEBRTC_SPL_MIN(tempMax, WEBRTC_SPL_WORD16_MAX);
+    return totMax;
+}
+
+#endif
+
+// Index of maximum absolute value in a  word16 vector.
+WebRtc_Word16 WebRtcSpl_MaxAbsIndexW16(G_CONST WebRtc_Word16* vector, WebRtc_Word16 length)
+{
+    WebRtc_Word16 tempMax;
+    WebRtc_Word16 absTemp;
+    WebRtc_Word16 tempMaxIndex = 0;
+    WebRtc_Word16 i = 0;
+    G_CONST WebRtc_Word16 *tmpvector = vector;
+
+    tempMax = WEBRTC_SPL_ABS_W16(*tmpvector);
+    tmpvector++;
+    for (i = 1; i < length; i++)
+    {
+        absTemp = WEBRTC_SPL_ABS_W16(*tmpvector);
+        tmpvector++;
+        if (absTemp > tempMax)
+        {
+            tempMax = absTemp;
+            tempMaxIndex = i;
+        }
+    }
+    return tempMaxIndex;
+}
+
+// Maximum absolute value of word32 vector.
+WebRtc_Word32 WebRtcSpl_MaxAbsValueW32(G_CONST WebRtc_Word32 *vector, WebRtc_Word16 length)
+{
+    WebRtc_UWord32 tempMax = 0;
+    WebRtc_UWord32 absVal;
+    WebRtc_Word32 retval;
+    int i;
+    G_CONST WebRtc_Word32 *tmpvector = vector;
+
+    for (i = 0; i < length; i++)
+    {
+        absVal = WEBRTC_SPL_ABS_W32((*tmpvector));
+        if (absVal > tempMax)
+        {
+            tempMax = absVal;
+        }
+        tmpvector++;
+    }
+    retval = (WebRtc_Word32)(WEBRTC_SPL_MIN(tempMax, WEBRTC_SPL_WORD32_MAX));
+    return retval;
+}
+
+// Maximum value of word16 vector.
+#ifndef XSCALE_OPT
+WebRtc_Word16 WebRtcSpl_MaxValueW16(G_CONST WebRtc_Word16* vector, WebRtc_Word16 length)
+{
+    WebRtc_Word16 tempMax;
+    WebRtc_Word16 i;
+    G_CONST WebRtc_Word16 *tmpvector = vector;
+
+    tempMax = *tmpvector++;
+    for (i = 1; i < length; i++)
+    {
+        if (*tmpvector++ > tempMax)
+            tempMax = vector[i];
+    }
+    return tempMax;
+}
+#else
+#pragma message(">> WebRtcSpl_MaxValueW16 is excluded from this build")
+#endif
+
+// Index of maximum value in a word16 vector.
+WebRtc_Word16 WebRtcSpl_MaxIndexW16(G_CONST WebRtc_Word16 *vector, WebRtc_Word16 length)
+{
+    WebRtc_Word16 tempMax;
+    WebRtc_Word16 tempMaxIndex = 0;
+    WebRtc_Word16 i = 0;
+    G_CONST WebRtc_Word16 *tmpvector = vector;
+
+    tempMax = *tmpvector++;
+    for (i = 1; i < length; i++)
+    {
+        if (*tmpvector++ > tempMax)
+        {
+            tempMax = vector[i];
+            tempMaxIndex = i;
+        }
+    }
+    return tempMaxIndex;
+}
+
+// Maximum value of word32 vector.
+#ifndef XSCALE_OPT
+WebRtc_Word32 WebRtcSpl_MaxValueW32(G_CONST WebRtc_Word32* vector, WebRtc_Word16 length)
+{
+    WebRtc_Word32 tempMax;
+    WebRtc_Word16 i;
+    G_CONST WebRtc_Word32 *tmpvector = vector;
+
+    tempMax = *tmpvector++;
+    for (i = 1; i < length; i++)
+    {
+        if (*tmpvector++ > tempMax)
+            tempMax = vector[i];
+    }
+    return tempMax;
+}
+#else
+#pragma message(">> WebRtcSpl_MaxValueW32 is excluded from this build")
+#endif
+
+// Index of maximum value in a word32 vector.
+WebRtc_Word16 WebRtcSpl_MaxIndexW32(G_CONST WebRtc_Word32* vector, WebRtc_Word16 length)
+{
+    WebRtc_Word32 tempMax;
+    WebRtc_Word16 tempMaxIndex = 0;
+    WebRtc_Word16 i = 0;
+    G_CONST WebRtc_Word32 *tmpvector = vector;
+
+    tempMax = *tmpvector++;
+    for (i = 1; i < length; i++)
+    {
+        if (*tmpvector++ > tempMax)
+        {
+            tempMax = vector[i];
+            tempMaxIndex = i;
+        }
+    }
+    return tempMaxIndex;
+}
+
+// Minimum value of word16 vector.
+WebRtc_Word16 WebRtcSpl_MinValueW16(G_CONST WebRtc_Word16 *vector, WebRtc_Word16 length)
+{
+    WebRtc_Word16 tempMin;
+    WebRtc_Word16 i;
+    G_CONST WebRtc_Word16 *tmpvector = vector;
+
+    // Find the minimum value
+    tempMin = *tmpvector++;
+    for (i = 1; i < length; i++)
+    {
+        if (*tmpvector++ < tempMin)
+            tempMin = (vector[i]);
+    }
+    return tempMin;
+}
+
+// Index of minimum value in a word16 vector.
+#ifndef XSCALE_OPT
+WebRtc_Word16 WebRtcSpl_MinIndexW16(G_CONST WebRtc_Word16* vector, WebRtc_Word16 length)
+{
+    WebRtc_Word16 tempMin;
+    WebRtc_Word16 tempMinIndex = 0;
+    WebRtc_Word16 i = 0;
+    G_CONST WebRtc_Word16* tmpvector = vector;
+
+    // Find index of smallest value
+    tempMin = *tmpvector++;
+    for (i = 1; i < length; i++)
+    {
+        if (*tmpvector++ < tempMin)
+        {
+            tempMin = vector[i];
+            tempMinIndex = i;
+        }
+    }
+    return tempMinIndex;
+}
+#else
+#pragma message(">> WebRtcSpl_MinIndexW16 is excluded from this build")
+#endif
+
+// Minimum value of word32 vector.
+WebRtc_Word32 WebRtcSpl_MinValueW32(G_CONST WebRtc_Word32 *vector, WebRtc_Word16 length)
+{
+    WebRtc_Word32 tempMin;
+    WebRtc_Word16 i;
+    G_CONST WebRtc_Word32 *tmpvector = vector;
+
+    // Find the minimum value
+    tempMin = *tmpvector++;
+    for (i = 1; i < length; i++)
+    {
+        if (*tmpvector++ < tempMin)
+            tempMin = (vector[i]);
+    }
+    return tempMin;
+}
+
+// Index of minimum value in a word32 vector.
+#ifndef XSCALE_OPT
+WebRtc_Word16 WebRtcSpl_MinIndexW32(G_CONST WebRtc_Word32* vector, WebRtc_Word16 length)
+{
+    WebRtc_Word32 tempMin;
+    WebRtc_Word16 tempMinIndex = 0;
+    WebRtc_Word16 i = 0;
+    G_CONST WebRtc_Word32 *tmpvector = vector;
+
+    // Find index of smallest value
+    tempMin = *tmpvector++;
+    for (i = 1; i < length; i++)
+    {
+        if (*tmpvector++ < tempMin)
+        {
+            tempMin = vector[i];
+            tempMinIndex = i;
+        }
+    }
+    return tempMinIndex;
+}
+#else
+#pragma message(">> WebRtcSpl_MinIndexW32 is excluded from this build")
+#endif

src/common_audio/signal_processing_library/main/source/min_max_operations_neon.c

@@ -0,0 +1,47 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#if (defined(WEBRTC_ANDROID) && defined(WEBRTC_ARCH_ARM_NEON))
+
+#include <arm_neon.h>
+
+#include "signal_processing_library.h"
+
+// Maximum absolute value of word16 vector.
+WebRtc_Word16 WebRtcSpl_MaxAbsValueW16(const WebRtc_Word16* vector,
+                                       WebRtc_Word16 length) {
+  WebRtc_Word32 temp_max = 0;
+  WebRtc_Word32 abs_val;
+  WebRtc_Word16 tot_max;
+  int i;
+
+  __asm__("vmov.i16 d25, #0" : : : "d25");
+
+  for (i = 0; i < length - 7; i += 8) {
+    __asm__("vld1.16 {d26, d27}, [%0]" : : "r"(&vector[i]) : "q13");
+    __asm__("vabs.s16 q13, q13" : : : "q13");
+    __asm__("vpmax.s16 d26, d27" : : : "q13");
+    __asm__("vpmax.s16 d25, d26" : : : "d25", "d26");
+  }
+  __asm__("vpmax.s16 d25, d25" : : : "d25");
+  __asm__("vpmax.s16 d25, d25" : : : "d25");
+  __asm__("vmov.s16 %0, d25[0]" : "=r"(temp_max): : "d25");
+
+  for (; i < length; i++) {
+    abs_val = WEBRTC_SPL_ABS_W32((vector[i]));
+    if (abs_val > temp_max) {
+      temp_max = abs_val;
+    }
+  }
+  tot_max = (WebRtc_Word16)WEBRTC_SPL_MIN(temp_max, WEBRTC_SPL_WORD16_MAX);
+  return tot_max;
+}
+
+#endif

src/common_audio/signal_processing_library/main/source/randn_table.c

@@ -0,0 +1,85 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * Table with 512 samples from a normal distribution with mean 1 and std 1
+ * The values are shifted up 13 steps (multiplied by 8192)
+ */
+
+#include "signal_processing_library.h"
+
+WebRtc_Word16 WebRtcSpl_kRandNTable[] =
+{
+    9178,    -7260,       40,    10189,     4894,    -3531,   -13779,    14764,
+   -4008,    -8884,    -8990,     1008,     7368,     5184,     3251,    -5817,
+   -9786,     5963,     1770,     8066,    -7135,    10772,    -2298,     1361,
+    6484,     2241,    -8633,      792,      199,    -3344,     6553,   -10079,
+  -15040,       95,    11608,   -12469,    14161,    -4176,     2476,     6403,
+   13685,   -16005,     6646,     2239,    10916,    -3004,     -602,    -3141,
+    2142,    14144,    -5829,     5305,     8209,     4713,     2697,    -5112,
+   16092,    -1210,    -2891,    -6631,    -5360,   -11878,    -6781,    -2739,
+   -6392,      536,    10923,    10872,     5059,    -4748,    -7770,     5477,
+      38,    -1025,    -2892,     1638,     6304,    14375,   -11028,     1553,
+   -1565,    10762,     -393,     4040,     5257,    12310,     6554,    -4799,
+    4899,    -6354,     1603,    -1048,    -2220,     8247,     -186,    -8944,
+  -12004,     2332,     4801,    -4933,     6371,      131,     8614,    -5927,
+   -8287,   -22760,     4033,   -15162,     3385,     3246,     3153,    -5250,
+    3766,      784,     6494,      -62,     3531,    -1582,    15572,      662,
+   -3952,     -330,    -3196,      669,     7236,    -2678,    -6569,    23319,
+   -8645,     -741,    14830,   -15976,     4903,      315,   -11342,    10311,
+    1858,    -7777,     2145,     5436,     5677,     -113,   -10033,      826,
+   -1353,    17210,     7768,      986,    -1471,     8291,    -4982,     8207,
+  -14911,    -6255,    -2449,   -11881,    -7059,   -11703,    -4338,     8025,
+    7538,    -2823,   -12490,     9470,    -1613,    -2529,   -10092,    -7807,
+    9480,     6970,   -12844,     5123,     3532,     4816,     4803,    -8455,
+   -5045,    14032,    -4378,    -1643,     5756,   -11041,    -2732,   -16618,
+   -6430,   -18375,    -3320,     6098,     5131,    -4269,    -8840,     2482,
+   -7048,     1547,   -21890,    -6505,    -7414,     -424,   -11722,     7955,
+    1653,   -17299,     1823,      473,    -9232,     3337,     1111,      873,
+    4018,    -8982,     9889,     3531,   -11763,    -3799,     7373,    -4539,
+    3231,     7054,    -8537,     7616,     6244,    16635,      447,    -2915,
+   13967,      705,    -2669,    -1520,    -1771,   -16188,     5956,     5117,
+    6371,    -9936,    -1448,     2480,     5128,     7550,    -8130,     5236,
+    8213,    -6443,     7707,    -1950,   -13811,     7218,     7031,    -3883,
+      67,     5731,    -2874,    13480,    -3743,     9298,    -3280,     3552,
+   -4425,      -18,    -3785,    -9988,    -5357,     5477,   -11794,     2117,
+    1416,    -9935,     3376,      802,    -5079,    -8243,    12652,       66,
+    3653,    -2368,     6781,   -21895,    -7227,     2487,     7839,     -385,
+    6646,    -7016,    -4658,     5531,    -1705,      834,      129,     3694,
+   -1343,     2238,   -22640,    -6417,   -11139,    11301,    -2945,    -3494,
+   -5626,      185,    -3615,    -2041,    -7972,    -3106,      -60,   -23497,
+   -1566,    17064,     3519,     2518,      304,    -6805,   -10269,     2105,
+    1936,     -426,     -736,    -8122,    -1467,     4238,    -6939,   -13309,
+     360,     7402,    -7970,    12576,     3287,    12194,    -6289,   -16006,
+    9171,     4042,    -9193,     9123,    -2512,     6388,    -4734,    -8739,
+    1028,    -5406,    -1696,     5889,     -666,    -4736,     4971,     3565,
+    9362,    -6292,     3876,    -3652,   -19666,     7523,    -4061,      391,
+  -11773,     7502,    -3763,     4929,    -9478,    13278,     2805,     4496,
+    7814,    16419,    12455,   -14773,     2127,    -2746,     3763,     4847,
+    3698,     6978,     4751,    -6957,    -3581,      -45,     6252,     1513,
+   -4797,    -7925,    11270,    16188,    -2359,    -5269,     9376,   -10777,
+    7262,    20031,    -6515,    -2208,    -5353,     8085,    -1341,    -1303,
+    7333,     5576,     3625,     5763,    -7931,     9833,    -3371,   -10305,
+    6534,   -13539,    -9971,      997,     8464,    -4064,    -1495,     1857,
+   13624,     5458,     9490,   -11086,    -4524,    12022,     -550,     -198,
+     408,    -8455,    -7068,    10289,     9712,    -3366,     9028,    -7621,
+   -5243,     2362,     6909,     4672,    -4933,    -1799,     4709,    -4563,
+     -62,     -566,     1624,    -7010,    14730,   -17791,    -3697,    -2344,
+   -1741,     7099,    -9509,    -6855,    -1989,     3495,    -2289,     2031,
+   12784,      891,    14189,    -3963,    -5683,      421,   -12575,     1724,
+  -12682,    -5970,    -8169,     3143,    -1824,    -5488,    -5130,     8536,
+   12799,      794,     5738,     3459,   -11689,     -258,    -3738,    -3775,
+   -8742,     2333,     8312,    -9383,    10331,    13119,     8398,    10644,
+  -19433,    -6446,   -16277,   -11793,    16284,     9345,    15222,    15834,
+    2009,    -7349,      130,   -14547,      338,    -5998,     3337,    21492,
+    2406,     7703,     -951,    11196,     -564,     3406,     2217,     4806,
+    2374,    -5797,    11839,     8940,   -11874,    18213,     2855,    10492
+};

src/common_audio/signal_processing_library/main/source/randomization_functions.c

@@ -0,0 +1,52 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains implementations of the randomization functions
+ * WebRtcSpl_IncreaseSeed()
+ * WebRtcSpl_RandU()
+ * WebRtcSpl_RandN()
+ * WebRtcSpl_RandUArray()
+ *
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+WebRtc_UWord32 WebRtcSpl_IncreaseSeed(WebRtc_UWord32 *seed)
+{
+    seed[0] = (seed[0] * ((WebRtc_Word32)69069) + 1) & (WEBRTC_SPL_MAX_SEED_USED - 1);
+    return seed[0];
+}
+
+WebRtc_Word16 WebRtcSpl_RandU(WebRtc_UWord32 *seed)
+{
+    return (WebRtc_Word16)(WebRtcSpl_IncreaseSeed(seed) >> 16);
+}
+
+WebRtc_Word16 WebRtcSpl_RandN(WebRtc_UWord32 *seed)
+{
+    return WebRtcSpl_kRandNTable[WebRtcSpl_IncreaseSeed(seed) >> 23];
+}
+
+// Creates an array of uniformly distributed variables
+WebRtc_Word16 WebRtcSpl_RandUArray(WebRtc_Word16* vector,
+                                   WebRtc_Word16 vector_length,
+                                   WebRtc_UWord32* seed)
+{
+    int i;
+    for (i = 0; i < vector_length; i++)
+    {
+        vector[i] = WebRtcSpl_RandU(seed);
+    }
+    return vector_length;
+}

src/common_audio/signal_processing_library/main/source/refl_coef_to_lpc.c

@@ -0,0 +1,60 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_ReflCoefToLpc().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+void WebRtcSpl_ReflCoefToLpc(G_CONST WebRtc_Word16 *k, int use_order, WebRtc_Word16 *a)
+{
+    WebRtc_Word16 any[WEBRTC_SPL_MAX_LPC_ORDER + 1];
+    WebRtc_Word16 *aptr, *aptr2, *anyptr;
+    G_CONST WebRtc_Word16 *kptr;
+    int m, i;
+
+    kptr = k;
+    *a = 4096; // i.e., (Word16_MAX >> 3)+1.
+    *any = *a;
+    a[1] = WEBRTC_SPL_RSHIFT_W16((*k), 3);
+
+    for (m = 1; m < use_order; m++)
+    {
+        kptr++;
+        aptr = a;
+        aptr++;
+        aptr2 = &a[m];
+        anyptr = any;
+        anyptr++;
+
+        any[m + 1] = WEBRTC_SPL_RSHIFT_W16((*kptr), 3);
+        for (i = 0; i < m; i++)
+        {
+            *anyptr = (*aptr)
+                    + (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT((*aptr2), (*kptr), 15);
+            anyptr++;
+            aptr++;
+            aptr2--;
+        }
+
+        aptr = a;
+        anyptr = any;
+        for (i = 0; i < (m + 2); i++)
+        {
+            *aptr = *anyptr;
+            aptr++;
+            anyptr++;
+        }
+    }
+}

src/common_audio/signal_processing_library/main/source/resample.c

@@ -0,0 +1,505 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the resampling functions for 22 kHz.
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+#include "resample_by_2_internal.h"
+
+// Declaration of internally used functions
+static void WebRtcSpl_32khzTo22khzIntToShort(const WebRtc_Word32 *In, WebRtc_Word16 *Out,
+                                             const WebRtc_Word32 K);
+
+void WebRtcSpl_32khzTo22khzIntToInt(const WebRtc_Word32 *In, WebRtc_Word32 *Out,
+                                    const WebRtc_Word32 K);
+
+// interpolation coefficients
+static const WebRtc_Word16 kCoefficients32To22[5][9] = {
+        {127, -712,  2359, -6333, 23456, 16775, -3695,  945, -154},
+        {-39,  230,  -830,  2785, 32366, -2324,   760, -218,   38},
+        {117, -663,  2222, -6133, 26634, 13070, -3174,  831, -137},
+        {-77,  457, -1677,  5958, 31175, -4136,  1405, -408,   71},
+        { 98, -560,  1900, -5406, 29240,  9423, -2480,  663, -110}
+};
+
+//////////////////////
+// 22 kHz -> 16 kHz //
+//////////////////////
+
+// number of subblocks; options: 1, 2, 4, 5, 10
+#define SUB_BLOCKS_22_16    5
+
+// 22 -> 16 resampler
+void WebRtcSpl_Resample22khzTo16khz(const WebRtc_Word16* in, WebRtc_Word16* out,
+                                    WebRtcSpl_State22khzTo16khz* state, WebRtc_Word32* tmpmem)
+{
+    int k;
+
+    // process two blocks of 10/SUB_BLOCKS_22_16 ms (to reduce temp buffer size)
+    for (k = 0; k < SUB_BLOCKS_22_16; k++)
+    {
+        ///// 22 --> 44 /////
+        // WebRtc_Word16  in[220/SUB_BLOCKS_22_16]
+        // WebRtc_Word32 out[440/SUB_BLOCKS_22_16]
+        /////
+        WebRtcSpl_UpBy2ShortToInt(in, 220 / SUB_BLOCKS_22_16, tmpmem + 16, state->S_22_44);
+
+        ///// 44 --> 32 /////
+        // WebRtc_Word32  in[440/SUB_BLOCKS_22_16]
+        // WebRtc_Word32 out[320/SUB_BLOCKS_22_16]
+        /////
+        // copy state to and from input array
+        tmpmem[8] = state->S_44_32[0];
+        tmpmem[9] = state->S_44_32[1];
+        tmpmem[10] = state->S_44_32[2];
+        tmpmem[11] = state->S_44_32[3];
+        tmpmem[12] = state->S_44_32[4];
+        tmpmem[13] = state->S_44_32[5];
+        tmpmem[14] = state->S_44_32[6];
+        tmpmem[15] = state->S_44_32[7];
+        state->S_44_32[0] = tmpmem[440 / SUB_BLOCKS_22_16 + 8];
+        state->S_44_32[1] = tmpmem[440 / SUB_BLOCKS_22_16 + 9];
+        state->S_44_32[2] = tmpmem[440 / SUB_BLOCKS_22_16 + 10];
+        state->S_44_32[3] = tmpmem[440 / SUB_BLOCKS_22_16 + 11];
+        state->S_44_32[4] = tmpmem[440 / SUB_BLOCKS_22_16 + 12];
+        state->S_44_32[5] = tmpmem[440 / SUB_BLOCKS_22_16 + 13];
+        state->S_44_32[6] = tmpmem[440 / SUB_BLOCKS_22_16 + 14];
+        state->S_44_32[7] = tmpmem[440 / SUB_BLOCKS_22_16 + 15];
+
+        WebRtcSpl_Resample44khzTo32khz(tmpmem + 8, tmpmem, 40 / SUB_BLOCKS_22_16);
+
+        ///// 32 --> 16 /////
+        // WebRtc_Word32  in[320/SUB_BLOCKS_22_16]
+        // WebRtc_Word32 out[160/SUB_BLOCKS_22_16]
+        /////
+        WebRtcSpl_DownBy2IntToShort(tmpmem, 320 / SUB_BLOCKS_22_16, out, state->S_32_16);
+
+        // move input/output pointers 10/SUB_BLOCKS_22_16 ms seconds ahead
+        in += 220 / SUB_BLOCKS_22_16;
+        out += 160 / SUB_BLOCKS_22_16;
+    }
+}
+
+// initialize state of 22 -> 16 resampler
+void WebRtcSpl_ResetResample22khzTo16khz(WebRtcSpl_State22khzTo16khz* state)
+{
+    int k;
+    for (k = 0; k < 8; k++)
+    {
+        state->S_22_44[k] = 0;
+        state->S_44_32[k] = 0;
+        state->S_32_16[k] = 0;
+    }
+}
+
+//////////////////////
+// 16 kHz -> 22 kHz //
+//////////////////////
+
+// number of subblocks; options: 1, 2, 4, 5, 10
+#define SUB_BLOCKS_16_22    4
+
+// 16 -> 22 resampler
+void WebRtcSpl_Resample16khzTo22khz(const WebRtc_Word16* in, WebRtc_Word16* out,
+                                    WebRtcSpl_State16khzTo22khz* state, WebRtc_Word32* tmpmem)
+{
+    int k;
+
+    // process two blocks of 10/SUB_BLOCKS_16_22 ms (to reduce temp buffer size)
+    for (k = 0; k < SUB_BLOCKS_16_22; k++)
+    {
+        ///// 16 --> 32 /////
+        // WebRtc_Word16  in[160/SUB_BLOCKS_16_22]
+        // WebRtc_Word32 out[320/SUB_BLOCKS_16_22]
+        /////
+        WebRtcSpl_UpBy2ShortToInt(in, 160 / SUB_BLOCKS_16_22, tmpmem + 8, state->S_16_32);
+
+        ///// 32 --> 22 /////
+        // WebRtc_Word32  in[320/SUB_BLOCKS_16_22]
+        // WebRtc_Word32 out[220/SUB_BLOCKS_16_22]
+        /////
+        // copy state to and from input array
+        tmpmem[0] = state->S_32_22[0];
+        tmpmem[1] = state->S_32_22[1];
+        tmpmem[2] = state->S_32_22[2];
+        tmpmem[3] = state->S_32_22[3];
+        tmpmem[4] = state->S_32_22[4];
+        tmpmem[5] = state->S_32_22[5];
+        tmpmem[6] = state->S_32_22[6];
+        tmpmem[7] = state->S_32_22[7];
+        state->S_32_22[0] = tmpmem[320 / SUB_BLOCKS_16_22];
+        state->S_32_22[1] = tmpmem[320 / SUB_BLOCKS_16_22 + 1];
+        state->S_32_22[2] = tmpmem[320 / SUB_BLOCKS_16_22 + 2];
+        state->S_32_22[3] = tmpmem[320 / SUB_BLOCKS_16_22 + 3];
+        state->S_32_22[4] = tmpmem[320 / SUB_BLOCKS_16_22 + 4];
+        state->S_32_22[5] = tmpmem[320 / SUB_BLOCKS_16_22 + 5];
+        state->S_32_22[6] = tmpmem[320 / SUB_BLOCKS_16_22 + 6];
+        state->S_32_22[7] = tmpmem[320 / SUB_BLOCKS_16_22 + 7];
+
+        WebRtcSpl_32khzTo22khzIntToShort(tmpmem, out, 20 / SUB_BLOCKS_16_22);
+
+        // move input/output pointers 10/SUB_BLOCKS_16_22 ms seconds ahead
+        in += 160 / SUB_BLOCKS_16_22;
+        out += 220 / SUB_BLOCKS_16_22;
+    }
+}
+
+// initialize state of 16 -> 22 resampler
+void WebRtcSpl_ResetResample16khzTo22khz(WebRtcSpl_State16khzTo22khz* state)
+{
+    int k;
+    for (k = 0; k < 8; k++)
+    {
+        state->S_16_32[k] = 0;
+        state->S_32_22[k] = 0;
+    }
+}
+
+//////////////////////
+// 22 kHz ->  8 kHz //
+//////////////////////
+
+// number of subblocks; options: 1, 2, 5, 10
+#define SUB_BLOCKS_22_8     2
+
+// 22 -> 8 resampler
+void WebRtcSpl_Resample22khzTo8khz(const WebRtc_Word16* in, WebRtc_Word16* out,
+                                   WebRtcSpl_State22khzTo8khz* state, WebRtc_Word32* tmpmem)
+{
+    int k;
+
+    // process two blocks of 10/SUB_BLOCKS_22_8 ms (to reduce temp buffer size)
+    for (k = 0; k < SUB_BLOCKS_22_8; k++)
+    {
+        ///// 22 --> 22 lowpass /////
+        // WebRtc_Word16  in[220/SUB_BLOCKS_22_8]
+        // WebRtc_Word32 out[220/SUB_BLOCKS_22_8]
+        /////
+        WebRtcSpl_LPBy2ShortToInt(in, 220 / SUB_BLOCKS_22_8, tmpmem + 16, state->S_22_22);
+
+        ///// 22 --> 16 /////
+        // WebRtc_Word32  in[220/SUB_BLOCKS_22_8]
+        // WebRtc_Word32 out[160/SUB_BLOCKS_22_8]
+        /////
+        // copy state to and from input array
+        tmpmem[8] = state->S_22_16[0];
+        tmpmem[9] = state->S_22_16[1];
+        tmpmem[10] = state->S_22_16[2];
+        tmpmem[11] = state->S_22_16[3];
+        tmpmem[12] = state->S_22_16[4];
+        tmpmem[13] = state->S_22_16[5];
+        tmpmem[14] = state->S_22_16[6];
+        tmpmem[15] = state->S_22_16[7];
+        state->S_22_16[0] = tmpmem[220 / SUB_BLOCKS_22_8 + 8];
+        state->S_22_16[1] = tmpmem[220 / SUB_BLOCKS_22_8 + 9];
+        state->S_22_16[2] = tmpmem[220 / SUB_BLOCKS_22_8 + 10];
+        state->S_22_16[3] = tmpmem[220 / SUB_BLOCKS_22_8 + 11];
+        state->S_22_16[4] = tmpmem[220 / SUB_BLOCKS_22_8 + 12];
+        state->S_22_16[5] = tmpmem[220 / SUB_BLOCKS_22_8 + 13];
+        state->S_22_16[6] = tmpmem[220 / SUB_BLOCKS_22_8 + 14];
+        state->S_22_16[7] = tmpmem[220 / SUB_BLOCKS_22_8 + 15];
+
+        WebRtcSpl_Resample44khzTo32khz(tmpmem + 8, tmpmem, 20 / SUB_BLOCKS_22_8);
+
+        ///// 16 --> 8 /////
+        // WebRtc_Word32 in[160/SUB_BLOCKS_22_8]
+        // WebRtc_Word32 out[80/SUB_BLOCKS_22_8]
+        /////
+        WebRtcSpl_DownBy2IntToShort(tmpmem, 160 / SUB_BLOCKS_22_8, out, state->S_16_8);
+
+        // move input/output pointers 10/SUB_BLOCKS_22_8 ms seconds ahead
+        in += 220 / SUB_BLOCKS_22_8;
+        out += 80 / SUB_BLOCKS_22_8;
+    }
+}
+
+// initialize state of 22 -> 8 resampler
+void WebRtcSpl_ResetResample22khzTo8khz(WebRtcSpl_State22khzTo8khz* state)
+{
+    int k;
+    for (k = 0; k < 8; k++)
+    {
+        state->S_22_22[k] = 0;
+        state->S_22_22[k + 8] = 0;
+        state->S_22_16[k] = 0;
+        state->S_16_8[k] = 0;
+    }
+}
+
+//////////////////////
+//  8 kHz -> 22 kHz //
+//////////////////////
+
+// number of subblocks; options: 1, 2, 5, 10
+#define SUB_BLOCKS_8_22     2
+
+// 8 -> 22 resampler
+void WebRtcSpl_Resample8khzTo22khz(const WebRtc_Word16* in, WebRtc_Word16* out,
+                                   WebRtcSpl_State8khzTo22khz* state, WebRtc_Word32* tmpmem)
+{
+    int k;
+
+    // process two blocks of 10/SUB_BLOCKS_8_22 ms (to reduce temp buffer size)
+    for (k = 0; k < SUB_BLOCKS_8_22; k++)
+    {
+        ///// 8 --> 16 /////
+        // WebRtc_Word16  in[80/SUB_BLOCKS_8_22]
+        // WebRtc_Word32 out[160/SUB_BLOCKS_8_22]
+        /////
+        WebRtcSpl_UpBy2ShortToInt(in, 80 / SUB_BLOCKS_8_22, tmpmem + 18, state->S_8_16);
+
+        ///// 16 --> 11 /////
+        // WebRtc_Word32  in[160/SUB_BLOCKS_8_22]
+        // WebRtc_Word32 out[110/SUB_BLOCKS_8_22]
+        /////
+        // copy state to and from input array
+        tmpmem[10] = state->S_16_11[0];
+        tmpmem[11] = state->S_16_11[1];
+        tmpmem[12] = state->S_16_11[2];
+        tmpmem[13] = state->S_16_11[3];
+        tmpmem[14] = state->S_16_11[4];
+        tmpmem[15] = state->S_16_11[5];
+        tmpmem[16] = state->S_16_11[6];
+        tmpmem[17] = state->S_16_11[7];
+        state->S_16_11[0] = tmpmem[160 / SUB_BLOCKS_8_22 + 10];
+        state->S_16_11[1] = tmpmem[160 / SUB_BLOCKS_8_22 + 11];
+        state->S_16_11[2] = tmpmem[160 / SUB_BLOCKS_8_22 + 12];
+        state->S_16_11[3] = tmpmem[160 / SUB_BLOCKS_8_22 + 13];
+        state->S_16_11[4] = tmpmem[160 / SUB_BLOCKS_8_22 + 14];
+        state->S_16_11[5] = tmpmem[160 / SUB_BLOCKS_8_22 + 15];
+        state->S_16_11[6] = tmpmem[160 / SUB_BLOCKS_8_22 + 16];
+        state->S_16_11[7] = tmpmem[160 / SUB_BLOCKS_8_22 + 17];
+
+        WebRtcSpl_32khzTo22khzIntToInt(tmpmem + 10, tmpmem, 10 / SUB_BLOCKS_8_22);
+
+        ///// 11 --> 22 /////
+        // WebRtc_Word32  in[110/SUB_BLOCKS_8_22]
+        // WebRtc_Word16 out[220/SUB_BLOCKS_8_22]
+        /////
+        WebRtcSpl_UpBy2IntToShort(tmpmem, 110 / SUB_BLOCKS_8_22, out, state->S_11_22);
+
+        // move input/output pointers 10/SUB_BLOCKS_8_22 ms seconds ahead
+        in += 80 / SUB_BLOCKS_8_22;
+        out += 220 / SUB_BLOCKS_8_22;
+    }
+}
+
+// initialize state of 8 -> 22 resampler
+void WebRtcSpl_ResetResample8khzTo22khz(WebRtcSpl_State8khzTo22khz* state)
+{
+    int k;
+    for (k = 0; k < 8; k++)
+    {
+        state->S_8_16[k] = 0;
+        state->S_16_11[k] = 0;
+        state->S_11_22[k] = 0;
+    }
+}
+
+// compute two inner-products and store them to output array
+static void WebRtcSpl_DotProdIntToInt(const WebRtc_Word32* in1, const WebRtc_Word32* in2,
+                                      const WebRtc_Word16* coef_ptr, WebRtc_Word32* out1,
+                                      WebRtc_Word32* out2)
+{
+    WebRtc_Word32 tmp1 = 16384;
+    WebRtc_Word32 tmp2 = 16384;
+    WebRtc_Word16 coef;
+
+    coef = coef_ptr[0];
+    tmp1 += coef * in1[0];
+    tmp2 += coef * in2[-0];
+
+    coef = coef_ptr[1];
+    tmp1 += coef * in1[1];
+    tmp2 += coef * in2[-1];
+
+    coef = coef_ptr[2];
+    tmp1 += coef * in1[2];
+    tmp2 += coef * in2[-2];
+
+    coef = coef_ptr[3];
+    tmp1 += coef * in1[3];
+    tmp2 += coef * in2[-3];
+
+    coef = coef_ptr[4];
+    tmp1 += coef * in1[4];
+    tmp2 += coef * in2[-4];
+
+    coef = coef_ptr[5];
+    tmp1 += coef * in1[5];
+    tmp2 += coef * in2[-5];
+
+    coef = coef_ptr[6];
+    tmp1 += coef * in1[6];
+    tmp2 += coef * in2[-6];
+
+    coef = coef_ptr[7];
+    tmp1 += coef * in1[7];
+    tmp2 += coef * in2[-7];
+
+    coef = coef_ptr[8];
+    *out1 = tmp1 + coef * in1[8];
+    *out2 = tmp2 + coef * in2[-8];
+}
+
+// compute two inner-products and store them to output array
+static void WebRtcSpl_DotProdIntToShort(const WebRtc_Word32* in1, const WebRtc_Word32* in2,
+                                        const WebRtc_Word16* coef_ptr, WebRtc_Word16* out1,
+                                        WebRtc_Word16* out2)
+{
+    WebRtc_Word32 tmp1 = 16384;
+    WebRtc_Word32 tmp2 = 16384;
+    WebRtc_Word16 coef;
+
+    coef = coef_ptr[0];
+    tmp1 += coef * in1[0];
+    tmp2 += coef * in2[-0];
+
+    coef = coef_ptr[1];
+    tmp1 += coef * in1[1];
+    tmp2 += coef * in2[-1];
+
+    coef = coef_ptr[2];
+    tmp1 += coef * in1[2];
+    tmp2 += coef * in2[-2];
+
+    coef = coef_ptr[3];
+    tmp1 += coef * in1[3];
+    tmp2 += coef * in2[-3];
+
+    coef = coef_ptr[4];
+    tmp1 += coef * in1[4];
+    tmp2 += coef * in2[-4];
+
+    coef = coef_ptr[5];
+    tmp1 += coef * in1[5];
+    tmp2 += coef * in2[-5];
+
+    coef = coef_ptr[6];
+    tmp1 += coef * in1[6];
+    tmp2 += coef * in2[-6];
+
+    coef = coef_ptr[7];
+    tmp1 += coef * in1[7];
+    tmp2 += coef * in2[-7];
+
+    coef = coef_ptr[8];
+    tmp1 += coef * in1[8];
+    tmp2 += coef * in2[-8];
+
+    // scale down, round and saturate
+    tmp1 >>= 15;
+    if (tmp1 > (WebRtc_Word32)0x00007FFF)
+        tmp1 = 0x00007FFF;
+    if (tmp1 < (WebRtc_Word32)0xFFFF8000)
+        tmp1 = 0xFFFF8000;
+    tmp2 >>= 15;
+    if (tmp2 > (WebRtc_Word32)0x00007FFF)
+        tmp2 = 0x00007FFF;
+    if (tmp2 < (WebRtc_Word32)0xFFFF8000)
+        tmp2 = 0xFFFF8000;
+    *out1 = (WebRtc_Word16)tmp1;
+    *out2 = (WebRtc_Word16)tmp2;
+}
+
+//   Resampling ratio: 11/16
+// input:  WebRtc_Word32 (normalized, not saturated) :: size 16 * K
+// output: WebRtc_Word32 (shifted 15 positions to the left, + offset 16384) :: size 11 * K
+//      K: Number of blocks
+
+void WebRtcSpl_32khzTo22khzIntToInt(const WebRtc_Word32* In,
+                                    WebRtc_Word32* Out,
+                                    const WebRtc_Word32 K)
+{
+    /////////////////////////////////////////////////////////////
+    // Filter operation:
+    //
+    // Perform resampling (16 input samples -> 11 output samples);
+    // process in sub blocks of size 16 samples.
+    WebRtc_Word32 m;
+
+    for (m = 0; m < K; m++)
+    {
+        // first output sample
+        Out[0] = ((WebRtc_Word32)In[3] << 15) + (1 << 14);
+
+        // sum and accumulate filter coefficients and input samples
+        WebRtcSpl_DotProdIntToInt(&In[0], &In[22], kCoefficients32To22[0], &Out[1], &Out[10]);
+
+        // sum and accumulate filter coefficients and input samples
+        WebRtcSpl_DotProdIntToInt(&In[2], &In[20], kCoefficients32To22[1], &Out[2], &Out[9]);
+
+        // sum and accumulate filter coefficients and input samples
+        WebRtcSpl_DotProdIntToInt(&In[3], &In[19], kCoefficients32To22[2], &Out[3], &Out[8]);
+
+        // sum and accumulate filter coefficients and input samples
+        WebRtcSpl_DotProdIntToInt(&In[5], &In[17], kCoefficients32To22[3], &Out[4], &Out[7]);
+
+        // sum and accumulate filter coefficients and input samples
+        WebRtcSpl_DotProdIntToInt(&In[6], &In[16], kCoefficients32To22[4], &Out[5], &Out[6]);
+
+        // update pointers
+        In += 16;
+        Out += 11;
+    }
+}
+
+//   Resampling ratio: 11/16
+// input:  WebRtc_Word32 (normalized, not saturated) :: size 16 * K
+// output: WebRtc_Word16 (saturated) :: size 11 * K
+//      K: Number of blocks
+
+void WebRtcSpl_32khzTo22khzIntToShort(const WebRtc_Word32 *In,
+                                      WebRtc_Word16 *Out,
+                                      const WebRtc_Word32 K)
+{
+    /////////////////////////////////////////////////////////////
+    // Filter operation:
+    //
+    // Perform resampling (16 input samples -> 11 output samples);
+    // process in sub blocks of size 16 samples.
+    WebRtc_Word32 tmp;
+    WebRtc_Word32 m;
+
+    for (m = 0; m < K; m++)
+    {
+        // first output sample
+        tmp = In[3];
+        if (tmp > (WebRtc_Word32)0x00007FFF)
+            tmp = 0x00007FFF;
+        if (tmp < (WebRtc_Word32)0xFFFF8000)
+            tmp = 0xFFFF8000;
+        Out[0] = (WebRtc_Word16)tmp;
+
+        // sum and accumulate filter coefficients and input samples
+        WebRtcSpl_DotProdIntToShort(&In[0], &In[22], kCoefficients32To22[0], &Out[1], &Out[10]);
+
+        // sum and accumulate filter coefficients and input samples
+        WebRtcSpl_DotProdIntToShort(&In[2], &In[20], kCoefficients32To22[1], &Out[2], &Out[9]);
+
+        // sum and accumulate filter coefficients and input samples
+        WebRtcSpl_DotProdIntToShort(&In[3], &In[19], kCoefficients32To22[2], &Out[3], &Out[8]);
+
+        // sum and accumulate filter coefficients and input samples
+        WebRtcSpl_DotProdIntToShort(&In[5], &In[17], kCoefficients32To22[3], &Out[4], &Out[7]);
+
+        // sum and accumulate filter coefficients and input samples
+        WebRtcSpl_DotProdIntToShort(&In[6], &In[16], kCoefficients32To22[4], &Out[5], &Out[6]);
+
+        // update pointers
+        In += 16;
+        Out += 11;
+    }
+}

src/common_audio/signal_processing_library/main/source/resample_48khz.c

@@ -0,0 +1,186 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains resampling functions between 48 kHz and nb/wb.
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include <string.h>
+#include "signal_processing_library.h"
+#include "resample_by_2_internal.h"
+
+////////////////////////////
+///// 48 kHz -> 16 kHz /////
+////////////////////////////
+
+// 48 -> 16 resampler
+void WebRtcSpl_Resample48khzTo16khz(const WebRtc_Word16* in, WebRtc_Word16* out,
+                                    WebRtcSpl_State48khzTo16khz* state, WebRtc_Word32* tmpmem)
+{
+    ///// 48 --> 48(LP) /////
+    // WebRtc_Word16  in[480]
+    // WebRtc_Word32 out[480]
+    /////
+    WebRtcSpl_LPBy2ShortToInt(in, 480, tmpmem + 16, state->S_48_48);
+
+    ///// 48 --> 32 /////
+    // WebRtc_Word32  in[480]
+    // WebRtc_Word32 out[320]
+    /////
+    // copy state to and from input array
+    memcpy(tmpmem + 8, state->S_48_32, 8 * sizeof(WebRtc_Word32));
+    memcpy(state->S_48_32, tmpmem + 488, 8 * sizeof(WebRtc_Word32));
+    WebRtcSpl_Resample48khzTo32khz(tmpmem + 8, tmpmem, 160);
+
+    ///// 32 --> 16 /////
+    // WebRtc_Word32  in[320]
+    // WebRtc_Word16 out[160]
+    /////
+    WebRtcSpl_DownBy2IntToShort(tmpmem, 320, out, state->S_32_16);
+}
+
+// initialize state of 48 -> 16 resampler
+void WebRtcSpl_ResetResample48khzTo16khz(WebRtcSpl_State48khzTo16khz* state)
+{
+    memset(state->S_48_48, 0, 16 * sizeof(WebRtc_Word32));
+    memset(state->S_48_32, 0, 8 * sizeof(WebRtc_Word32));
+    memset(state->S_32_16, 0, 8 * sizeof(WebRtc_Word32));
+}
+
+////////////////////////////
+///// 16 kHz -> 48 kHz /////
+////////////////////////////
+
+// 16 -> 48 resampler
+void WebRtcSpl_Resample16khzTo48khz(const WebRtc_Word16* in, WebRtc_Word16* out,
+                                    WebRtcSpl_State16khzTo48khz* state, WebRtc_Word32* tmpmem)
+{
+    ///// 16 --> 32 /////
+    // WebRtc_Word16  in[160]
+    // WebRtc_Word32 out[320]
+    /////
+    WebRtcSpl_UpBy2ShortToInt(in, 160, tmpmem + 16, state->S_16_32);
+
+    ///// 32 --> 24 /////
+    // WebRtc_Word32  in[320]
+    // WebRtc_Word32 out[240]
+    // copy state to and from input array
+    /////
+    memcpy(tmpmem + 8, state->S_32_24, 8 * sizeof(WebRtc_Word32));
+    memcpy(state->S_32_24, tmpmem + 328, 8 * sizeof(WebRtc_Word32));
+    WebRtcSpl_Resample32khzTo24khz(tmpmem + 8, tmpmem, 80);
+
+    ///// 24 --> 48 /////
+    // WebRtc_Word32  in[240]
+    // WebRtc_Word16 out[480]
+    /////
+    WebRtcSpl_UpBy2IntToShort(tmpmem, 240, out, state->S_24_48);
+}
+
+// initialize state of 16 -> 48 resampler
+void WebRtcSpl_ResetResample16khzTo48khz(WebRtcSpl_State16khzTo48khz* state)
+{
+    memset(state->S_16_32, 0, 8 * sizeof(WebRtc_Word32));
+    memset(state->S_32_24, 0, 8 * sizeof(WebRtc_Word32));
+    memset(state->S_24_48, 0, 8 * sizeof(WebRtc_Word32));
+}
+
+////////////////////////////
+///// 48 kHz ->  8 kHz /////
+////////////////////////////
+
+// 48 -> 8 resampler
+void WebRtcSpl_Resample48khzTo8khz(const WebRtc_Word16* in, WebRtc_Word16* out,
+                                   WebRtcSpl_State48khzTo8khz* state, WebRtc_Word32* tmpmem)
+{
+    ///// 48 --> 24 /////
+    // WebRtc_Word16  in[480]
+    // WebRtc_Word32 out[240]
+    /////
+    WebRtcSpl_DownBy2ShortToInt(in, 480, tmpmem + 256, state->S_48_24);
+
+    ///// 24 --> 24(LP) /////
+    // WebRtc_Word32  in[240]
+    // WebRtc_Word32 out[240]
+    /////
+    WebRtcSpl_LPBy2IntToInt(tmpmem + 256, 240, tmpmem + 16, state->S_24_24);
+
+    ///// 24 --> 16 /////
+    // WebRtc_Word32  in[240]
+    // WebRtc_Word32 out[160]
+    /////
+    // copy state to and from input array
+    memcpy(tmpmem + 8, state->S_24_16, 8 * sizeof(WebRtc_Word32));
+    memcpy(state->S_24_16, tmpmem + 248, 8 * sizeof(WebRtc_Word32));
+    WebRtcSpl_Resample48khzTo32khz(tmpmem + 8, tmpmem, 80);
+
+    ///// 16 --> 8 /////
+    // WebRtc_Word32  in[160]
+    // WebRtc_Word16 out[80]
+    /////
+    WebRtcSpl_DownBy2IntToShort(tmpmem, 160, out, state->S_16_8);
+}
+
+// initialize state of 48 -> 8 resampler
+void WebRtcSpl_ResetResample48khzTo8khz(WebRtcSpl_State48khzTo8khz* state)
+{
+    memset(state->S_48_24, 0, 8 * sizeof(WebRtc_Word32));
+    memset(state->S_24_24, 0, 16 * sizeof(WebRtc_Word32));
+    memset(state->S_24_16, 0, 8 * sizeof(WebRtc_Word32));
+    memset(state->S_16_8, 0, 8 * sizeof(WebRtc_Word32));
+}
+
+////////////////////////////
+/////  8 kHz -> 48 kHz /////
+////////////////////////////
+
+// 8 -> 48 resampler
+void WebRtcSpl_Resample8khzTo48khz(const WebRtc_Word16* in, WebRtc_Word16* out,
+                                   WebRtcSpl_State8khzTo48khz* state, WebRtc_Word32* tmpmem)
+{
+    ///// 8 --> 16 /////
+    // WebRtc_Word16  in[80]
+    // WebRtc_Word32 out[160]
+    /////
+    WebRtcSpl_UpBy2ShortToInt(in, 80, tmpmem + 264, state->S_8_16);
+
+    ///// 16 --> 12 /////
+    // WebRtc_Word32  in[160]
+    // WebRtc_Word32 out[120]
+    /////
+    // copy state to and from input array
+    memcpy(tmpmem + 256, state->S_16_12, 8 * sizeof(WebRtc_Word32));
+    memcpy(state->S_16_12, tmpmem + 416, 8 * sizeof(WebRtc_Word32));
+    WebRtcSpl_Resample32khzTo24khz(tmpmem + 256, tmpmem + 240, 40);
+
+    ///// 12 --> 24 /////
+    // WebRtc_Word32  in[120]
+    // WebRtc_Word16 out[240]
+    /////
+    WebRtcSpl_UpBy2IntToInt(tmpmem + 240, 120, tmpmem, state->S_12_24);
+
+    ///// 24 --> 48 /////
+    // WebRtc_Word32  in[240]
+    // WebRtc_Word16 out[480]
+    /////
+    WebRtcSpl_UpBy2IntToShort(tmpmem, 240, out, state->S_24_48);
+}
+
+// initialize state of 8 -> 48 resampler
+void WebRtcSpl_ResetResample8khzTo48khz(WebRtcSpl_State8khzTo48khz* state)
+{
+    memset(state->S_8_16, 0, 8 * sizeof(WebRtc_Word32));
+    memset(state->S_16_12, 0, 8 * sizeof(WebRtc_Word32));
+    memset(state->S_12_24, 0, 8 * sizeof(WebRtc_Word32));
+    memset(state->S_24_48, 0, 8 * sizeof(WebRtc_Word32));
+}

src/common_audio/signal_processing_library/main/source/resample_by_2.c

@@ -0,0 +1,170 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the resampling by two functions.
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+// allpass filter coefficients.
+static const WebRtc_UWord16 kResampleAllpass1[3] = {3284, 24441, 49528};
+static const WebRtc_UWord16 kResampleAllpass2[3] = {12199, 37471, 60255};
+
+// decimator
+void WebRtcSpl_DownsampleBy2(const WebRtc_Word16* in, const WebRtc_Word16 len,
+                             WebRtc_Word16* out, WebRtc_Word32* filtState)
+{
+    WebRtc_Word32 tmp1, tmp2, diff, in32, out32;
+    WebRtc_Word16 i;
+
+    register WebRtc_Word32 state0 = filtState[0];
+    register WebRtc_Word32 state1 = filtState[1];
+    register WebRtc_Word32 state2 = filtState[2];
+    register WebRtc_Word32 state3 = filtState[3];
+    register WebRtc_Word32 state4 = filtState[4];
+    register WebRtc_Word32 state5 = filtState[5];
+    register WebRtc_Word32 state6 = filtState[6];
+    register WebRtc_Word32 state7 = filtState[7];
+
+    for (i = (len >> 1); i > 0; i--)
+    {
+        // lower allpass filter
+        in32 = (WebRtc_Word32)(*in++) << 10;
+        diff = in32 - state1;
+        tmp1 = WEBRTC_SPL_SCALEDIFF32(kResampleAllpass2[0], diff, state0);
+        state0 = in32;
+        diff = tmp1 - state2;
+        tmp2 = WEBRTC_SPL_SCALEDIFF32(kResampleAllpass2[1], diff, state1);
+        state1 = tmp1;
+        diff = tmp2 - state3;
+        state3 = WEBRTC_SPL_SCALEDIFF32(kResampleAllpass2[2], diff, state2);
+        state2 = tmp2;
+
+        // upper allpass filter
+        in32 = (WebRtc_Word32)(*in++) << 10;
+        diff = in32 - state5;
+        tmp1 = WEBRTC_SPL_SCALEDIFF32(kResampleAllpass1[0], diff, state4);
+        state4 = in32;
+        diff = tmp1 - state6;
+        tmp2 = WEBRTC_SPL_SCALEDIFF32(kResampleAllpass1[1], diff, state5);
+        state5 = tmp1;
+        diff = tmp2 - state7;
+        state7 = WEBRTC_SPL_SCALEDIFF32(kResampleAllpass1[2], diff, state6);
+        state6 = tmp2;
+
+        // add two allpass outputs, divide by two and round
+        out32 = (state3 + state7 + 1024) >> 11;
+
+        // limit amplitude to prevent wrap-around, and write to output array
+#ifdef WEBRTC_ARCH_ARM_V7A
+        __asm__("ssat %r0, #16, %r1" : "=r"(*out) : "r"(out32));
+        out++;
+#else
+        if (out32 > 32767)
+            *out++ = 32767;
+        else if (out32 < -32768)
+            *out++ = -32768;
+        else
+            *out++ = (WebRtc_Word16)out32;
+#endif
+    }
+
+    filtState[0] = state0;
+    filtState[1] = state1;
+    filtState[2] = state2;
+    filtState[3] = state3;
+    filtState[4] = state4;
+    filtState[5] = state5;
+    filtState[6] = state6;
+    filtState[7] = state7;
+}
+
+void WebRtcSpl_UpsampleBy2(const WebRtc_Word16* in, WebRtc_Word16 len, WebRtc_Word16* out,
+                           WebRtc_Word32* filtState)
+{
+    WebRtc_Word32 tmp1, tmp2, diff, in32, out32;
+    WebRtc_Word16 i;
+
+    register WebRtc_Word32 state0 = filtState[0];
+    register WebRtc_Word32 state1 = filtState[1];
+    register WebRtc_Word32 state2 = filtState[2];
+    register WebRtc_Word32 state3 = filtState[3];
+    register WebRtc_Word32 state4 = filtState[4];
+    register WebRtc_Word32 state5 = filtState[5];
+    register WebRtc_Word32 state6 = filtState[6];
+    register WebRtc_Word32 state7 = filtState[7];
+
+    for (i = len; i > 0; i--)
+    {
+        // lower allpass filter
+        in32 = (WebRtc_Word32)(*in++) << 10;
+        diff = in32 - state1;
+        tmp1 = WEBRTC_SPL_SCALEDIFF32(kResampleAllpass1[0], diff, state0);
+        state0 = in32;
+        diff = tmp1 - state2;
+        tmp2 = WEBRTC_SPL_SCALEDIFF32(kResampleAllpass1[1], diff, state1);
+        state1 = tmp1;
+        diff = tmp2 - state3;
+        state3 = WEBRTC_SPL_SCALEDIFF32(kResampleAllpass1[2], diff, state2);
+        state2 = tmp2;
+
+        // round; limit amplitude to prevent wrap-around; write to output array
+        out32 = (state3 + 512) >> 10;
+#ifdef WEBRTC_ARCH_ARM_V7A
+        __asm__("ssat %r0, #16, %r1":"=r"(*out): "r"(out32));
+        out++;
+#else
+        if (out32 > 32767)
+            *out++ = 32767;
+        else if (out32 < -32768)
+            *out++ = -32768;
+        else
+            *out++ = (WebRtc_Word16)out32;
+#endif
+
+        // upper allpass filter
+        diff = in32 - state5;
+        tmp1 = WEBRTC_SPL_SCALEDIFF32(kResampleAllpass2[0], diff, state4);
+        state4 = in32;
+        diff = tmp1 - state6;
+        tmp2 = WEBRTC_SPL_SCALEDIFF32(kResampleAllpass2[1], diff, state5);
+        state5 = tmp1;
+        diff = tmp2 - state7;
+        state7 = WEBRTC_SPL_SCALEDIFF32(kResampleAllpass2[2], diff, state6);
+        state6 = tmp2;
+
+        // round; limit amplitude to prevent wrap-around; write to output array
+        out32 = (state7 + 512) >> 10;
+#ifdef WEBRTC_ARCH_ARM_V7A
+        __asm__("ssat %r0, #16, %r1":"=r"(*out): "r"(out32));
+        out++;
+#else
+        if (out32 > 32767)
+            *out++ = 32767;
+        else if (out32 < -32768)
+            *out++ = -32768;
+        else
+            *out++ = (WebRtc_Word16)out32;
+#endif
+    }
+    
+    filtState[0] = state0;
+    filtState[1] = state1;
+    filtState[2] = state2;
+    filtState[3] = state3;
+    filtState[4] = state4;
+    filtState[5] = state5;
+    filtState[6] = state6;
+    filtState[7] = state7;
+}

src/common_audio/signal_processing_library/main/source/resample_by_2_internal.c

@@ -0,0 +1,679 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This header file contains some internal resampling functions.
+ *
+ */
+
+#include "resample_by_2_internal.h"
+
+// allpass filter coefficients.
+static const WebRtc_Word16 kResampleAllpass[2][3] = {
+        {821, 6110, 12382},
+        {3050, 9368, 15063}
+};
+
+//
+//   decimator
+// input:  WebRtc_Word32 (shifted 15 positions to the left, + offset 16384) OVERWRITTEN!
+// output: WebRtc_Word16 (saturated) (of length len/2)
+// state:  filter state array; length = 8
+
+void WebRtcSpl_DownBy2IntToShort(WebRtc_Word32 *in, WebRtc_Word32 len, WebRtc_Word16 *out,
+                                 WebRtc_Word32 *state)
+{
+    WebRtc_Word32 tmp0, tmp1, diff;
+    WebRtc_Word32 i;
+
+    len >>= 1;
+
+    // lower allpass filter (operates on even input samples)
+    for (i = 0; i < len; i++)
+    {
+        tmp0 = in[i << 1];
+        diff = tmp0 - state[1];
+        // scale down and round
+        diff = (diff + (1 << 13)) >> 14;
+        tmp1 = state[0] + diff * kResampleAllpass[1][0];
+        state[0] = tmp0;
+        diff = tmp1 - state[2];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        tmp0 = state[1] + diff * kResampleAllpass[1][1];
+        state[1] = tmp1;
+        diff = tmp0 - state[3];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        state[3] = state[2] + diff * kResampleAllpass[1][2];
+        state[2] = tmp0;
+
+        // divide by two and store temporarily
+        in[i << 1] = (state[3] >> 1);
+    }
+
+    in++;
+
+    // upper allpass filter (operates on odd input samples)
+    for (i = 0; i < len; i++)
+    {
+        tmp0 = in[i << 1];
+        diff = tmp0 - state[5];
+        // scale down and round
+        diff = (diff + (1 << 13)) >> 14;
+        tmp1 = state[4] + diff * kResampleAllpass[0][0];
+        state[4] = tmp0;
+        diff = tmp1 - state[6];
+        // scale down and round
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        tmp0 = state[5] + diff * kResampleAllpass[0][1];
+        state[5] = tmp1;
+        diff = tmp0 - state[7];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        state[7] = state[6] + diff * kResampleAllpass[0][2];
+        state[6] = tmp0;
+
+        // divide by two and store temporarily
+        in[i << 1] = (state[7] >> 1);
+    }
+
+    in--;
+
+    // combine allpass outputs
+    for (i = 0; i < len; i += 2)
+    {
+        // divide by two, add both allpass outputs and round
+        tmp0 = (in[i << 1] + in[(i << 1) + 1]) >> 15;
+        tmp1 = (in[(i << 1) + 2] + in[(i << 1) + 3]) >> 15;
+        if (tmp0 > (WebRtc_Word32)0x00007FFF)
+            tmp0 = 0x00007FFF;
+        if (tmp0 < (WebRtc_Word32)0xFFFF8000)
+            tmp0 = 0xFFFF8000;
+        out[i] = (WebRtc_Word16)tmp0;
+        if (tmp1 > (WebRtc_Word32)0x00007FFF)
+            tmp1 = 0x00007FFF;
+        if (tmp1 < (WebRtc_Word32)0xFFFF8000)
+            tmp1 = 0xFFFF8000;
+        out[i + 1] = (WebRtc_Word16)tmp1;
+    }
+}
+
+//
+//   decimator
+// input:  WebRtc_Word16
+// output: WebRtc_Word32 (shifted 15 positions to the left, + offset 16384) (of length len/2)
+// state:  filter state array; length = 8
+
+void WebRtcSpl_DownBy2ShortToInt(const WebRtc_Word16 *in,
+                                  WebRtc_Word32 len,
+                                  WebRtc_Word32 *out,
+                                  WebRtc_Word32 *state)
+{
+    WebRtc_Word32 tmp0, tmp1, diff;
+    WebRtc_Word32 i;
+
+    len >>= 1;
+
+    // lower allpass filter (operates on even input samples)
+    for (i = 0; i < len; i++)
+    {
+        tmp0 = ((WebRtc_Word32)in[i << 1] << 15) + (1 << 14);
+        diff = tmp0 - state[1];
+        // scale down and round
+        diff = (diff + (1 << 13)) >> 14;
+        tmp1 = state[0] + diff * kResampleAllpass[1][0];
+        state[0] = tmp0;
+        diff = tmp1 - state[2];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        tmp0 = state[1] + diff * kResampleAllpass[1][1];
+        state[1] = tmp1;
+        diff = tmp0 - state[3];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        state[3] = state[2] + diff * kResampleAllpass[1][2];
+        state[2] = tmp0;
+
+        // divide by two and store temporarily
+        out[i] = (state[3] >> 1);
+    }
+
+    in++;
+
+    // upper allpass filter (operates on odd input samples)
+    for (i = 0; i < len; i++)
+    {
+        tmp0 = ((WebRtc_Word32)in[i << 1] << 15) + (1 << 14);
+        diff = tmp0 - state[5];
+        // scale down and round
+        diff = (diff + (1 << 13)) >> 14;
+        tmp1 = state[4] + diff * kResampleAllpass[0][0];
+        state[4] = tmp0;
+        diff = tmp1 - state[6];
+        // scale down and round
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        tmp0 = state[5] + diff * kResampleAllpass[0][1];
+        state[5] = tmp1;
+        diff = tmp0 - state[7];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        state[7] = state[6] + diff * kResampleAllpass[0][2];
+        state[6] = tmp0;
+
+        // divide by two and store temporarily
+        out[i] += (state[7] >> 1);
+    }
+
+    in--;
+}
+
+//
+//   interpolator
+// input:  WebRtc_Word16
+// output: WebRtc_Word32 (normalized, not saturated) (of length len*2)
+// state:  filter state array; length = 8
+void WebRtcSpl_UpBy2ShortToInt(const WebRtc_Word16 *in, WebRtc_Word32 len, WebRtc_Word32 *out,
+                               WebRtc_Word32 *state)
+{
+    WebRtc_Word32 tmp0, tmp1, diff;
+    WebRtc_Word32 i;
+
+    // upper allpass filter (generates odd output samples)
+    for (i = 0; i < len; i++)
+    {
+        tmp0 = ((WebRtc_Word32)in[i] << 15) + (1 << 14);
+        diff = tmp0 - state[5];
+        // scale down and round
+        diff = (diff + (1 << 13)) >> 14;
+        tmp1 = state[4] + diff * kResampleAllpass[0][0];
+        state[4] = tmp0;
+        diff = tmp1 - state[6];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        tmp0 = state[5] + diff * kResampleAllpass[0][1];
+        state[5] = tmp1;
+        diff = tmp0 - state[7];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        state[7] = state[6] + diff * kResampleAllpass[0][2];
+        state[6] = tmp0;
+
+        // scale down, round and store
+        out[i << 1] = state[7] >> 15;
+    }
+
+    out++;
+
+    // lower allpass filter (generates even output samples)
+    for (i = 0; i < len; i++)
+    {
+        tmp0 = ((WebRtc_Word32)in[i] << 15) + (1 << 14);
+        diff = tmp0 - state[1];
+        // scale down and round
+        diff = (diff + (1 << 13)) >> 14;
+        tmp1 = state[0] + diff * kResampleAllpass[1][0];
+        state[0] = tmp0;
+        diff = tmp1 - state[2];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        tmp0 = state[1] + diff * kResampleAllpass[1][1];
+        state[1] = tmp1;
+        diff = tmp0 - state[3];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        state[3] = state[2] + diff * kResampleAllpass[1][2];
+        state[2] = tmp0;
+
+        // scale down, round and store
+        out[i << 1] = state[3] >> 15;
+    }
+}
+
+//
+//   interpolator
+// input:  WebRtc_Word32 (shifted 15 positions to the left, + offset 16384)
+// output: WebRtc_Word32 (shifted 15 positions to the left, + offset 16384) (of length len*2)
+// state:  filter state array; length = 8
+void WebRtcSpl_UpBy2IntToInt(const WebRtc_Word32 *in, WebRtc_Word32 len, WebRtc_Word32 *out,
+                             WebRtc_Word32 *state)
+{
+    WebRtc_Word32 tmp0, tmp1, diff;
+    WebRtc_Word32 i;
+
+    // upper allpass filter (generates odd output samples)
+    for (i = 0; i < len; i++)
+    {
+        tmp0 = in[i];
+        diff = tmp0 - state[5];
+        // scale down and round
+        diff = (diff + (1 << 13)) >> 14;
+        tmp1 = state[4] + diff * kResampleAllpass[0][0];
+        state[4] = tmp0;
+        diff = tmp1 - state[6];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        tmp0 = state[5] + diff * kResampleAllpass[0][1];
+        state[5] = tmp1;
+        diff = tmp0 - state[7];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        state[7] = state[6] + diff * kResampleAllpass[0][2];
+        state[6] = tmp0;
+
+        // scale down, round and store
+        out[i << 1] = state[7];
+    }
+
+    out++;
+
+    // lower allpass filter (generates even output samples)
+    for (i = 0; i < len; i++)
+    {
+        tmp0 = in[i];
+        diff = tmp0 - state[1];
+        // scale down and round
+        diff = (diff + (1 << 13)) >> 14;
+        tmp1 = state[0] + diff * kResampleAllpass[1][0];
+        state[0] = tmp0;
+        diff = tmp1 - state[2];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        tmp0 = state[1] + diff * kResampleAllpass[1][1];
+        state[1] = tmp1;
+        diff = tmp0 - state[3];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        state[3] = state[2] + diff * kResampleAllpass[1][2];
+        state[2] = tmp0;
+
+        // scale down, round and store
+        out[i << 1] = state[3];
+    }
+}
+
+//
+//   interpolator
+// input:  WebRtc_Word32 (shifted 15 positions to the left, + offset 16384)
+// output: WebRtc_Word16 (saturated) (of length len*2)
+// state:  filter state array; length = 8
+void WebRtcSpl_UpBy2IntToShort(const WebRtc_Word32 *in, WebRtc_Word32 len, WebRtc_Word16 *out,
+                               WebRtc_Word32 *state)
+{
+    WebRtc_Word32 tmp0, tmp1, diff;
+    WebRtc_Word32 i;
+
+    // upper allpass filter (generates odd output samples)
+    for (i = 0; i < len; i++)
+    {
+        tmp0 = in[i];
+        diff = tmp0 - state[5];
+        // scale down and round
+        diff = (diff + (1 << 13)) >> 14;
+        tmp1 = state[4] + diff * kResampleAllpass[0][0];
+        state[4] = tmp0;
+        diff = tmp1 - state[6];
+        // scale down and round
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        tmp0 = state[5] + diff * kResampleAllpass[0][1];
+        state[5] = tmp1;
+        diff = tmp0 - state[7];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        state[7] = state[6] + diff * kResampleAllpass[0][2];
+        state[6] = tmp0;
+
+        // scale down, saturate and store
+        tmp1 = state[7] >> 15;
+        if (tmp1 > (WebRtc_Word32)0x00007FFF)
+            tmp1 = 0x00007FFF;
+        if (tmp1 < (WebRtc_Word32)0xFFFF8000)
+            tmp1 = 0xFFFF8000;
+        out[i << 1] = (WebRtc_Word16)tmp1;
+    }
+
+    out++;
+
+    // lower allpass filter (generates even output samples)
+    for (i = 0; i < len; i++)
+    {
+        tmp0 = in[i];
+        diff = tmp0 - state[1];
+        // scale down and round
+        diff = (diff + (1 << 13)) >> 14;
+        tmp1 = state[0] + diff * kResampleAllpass[1][0];
+        state[0] = tmp0;
+        diff = tmp1 - state[2];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        tmp0 = state[1] + diff * kResampleAllpass[1][1];
+        state[1] = tmp1;
+        diff = tmp0 - state[3];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        state[3] = state[2] + diff * kResampleAllpass[1][2];
+        state[2] = tmp0;
+
+        // scale down, saturate and store
+        tmp1 = state[3] >> 15;
+        if (tmp1 > (WebRtc_Word32)0x00007FFF)
+            tmp1 = 0x00007FFF;
+        if (tmp1 < (WebRtc_Word32)0xFFFF8000)
+            tmp1 = 0xFFFF8000;
+        out[i << 1] = (WebRtc_Word16)tmp1;
+    }
+}
+
+//   lowpass filter
+// input:  WebRtc_Word16
+// output: WebRtc_Word32 (normalized, not saturated)
+// state:  filter state array; length = 8
+void WebRtcSpl_LPBy2ShortToInt(const WebRtc_Word16* in, WebRtc_Word32 len, WebRtc_Word32* out,
+                               WebRtc_Word32* state)
+{
+    WebRtc_Word32 tmp0, tmp1, diff;
+    WebRtc_Word32 i;
+
+    len >>= 1;
+
+    // lower allpass filter: odd input -> even output samples
+    in++;
+    // initial state of polyphase delay element
+    tmp0 = state[12];
+    for (i = 0; i < len; i++)
+    {
+        diff = tmp0 - state[1];
+        // scale down and round
+        diff = (diff + (1 << 13)) >> 14;
+        tmp1 = state[0] + diff * kResampleAllpass[1][0];
+        state[0] = tmp0;
+        diff = tmp1 - state[2];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        tmp0 = state[1] + diff * kResampleAllpass[1][1];
+        state[1] = tmp1;
+        diff = tmp0 - state[3];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        state[3] = state[2] + diff * kResampleAllpass[1][2];
+        state[2] = tmp0;
+
+        // scale down, round and store
+        out[i << 1] = state[3] >> 1;
+        tmp0 = ((WebRtc_Word32)in[i << 1] << 15) + (1 << 14);
+    }
+    in--;
+
+    // upper allpass filter: even input -> even output samples
+    for (i = 0; i < len; i++)
+    {
+        tmp0 = ((WebRtc_Word32)in[i << 1] << 15) + (1 << 14);
+        diff = tmp0 - state[5];
+        // scale down and round
+        diff = (diff + (1 << 13)) >> 14;
+        tmp1 = state[4] + diff * kResampleAllpass[0][0];
+        state[4] = tmp0;
+        diff = tmp1 - state[6];
+        // scale down and round
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        tmp0 = state[5] + diff * kResampleAllpass[0][1];
+        state[5] = tmp1;
+        diff = tmp0 - state[7];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        state[7] = state[6] + diff * kResampleAllpass[0][2];
+        state[6] = tmp0;
+
+        // average the two allpass outputs, scale down and store
+        out[i << 1] = (out[i << 1] + (state[7] >> 1)) >> 15;
+    }
+
+    // switch to odd output samples
+    out++;
+
+    // lower allpass filter: even input -> odd output samples
+    for (i = 0; i < len; i++)
+    {
+        tmp0 = ((WebRtc_Word32)in[i << 1] << 15) + (1 << 14);
+        diff = tmp0 - state[9];
+        // scale down and round
+        diff = (diff + (1 << 13)) >> 14;
+        tmp1 = state[8] + diff * kResampleAllpass[1][0];
+        state[8] = tmp0;
+        diff = tmp1 - state[10];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        tmp0 = state[9] + diff * kResampleAllpass[1][1];
+        state[9] = tmp1;
+        diff = tmp0 - state[11];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        state[11] = state[10] + diff * kResampleAllpass[1][2];
+        state[10] = tmp0;
+
+        // scale down, round and store
+        out[i << 1] = state[11] >> 1;
+    }
+
+    // upper allpass filter: odd input -> odd output samples
+    in++;
+    for (i = 0; i < len; i++)
+    {
+        tmp0 = ((WebRtc_Word32)in[i << 1] << 15) + (1 << 14);
+        diff = tmp0 - state[13];
+        // scale down and round
+        diff = (diff + (1 << 13)) >> 14;
+        tmp1 = state[12] + diff * kResampleAllpass[0][0];
+        state[12] = tmp0;
+        diff = tmp1 - state[14];
+        // scale down and round
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        tmp0 = state[13] + diff * kResampleAllpass[0][1];
+        state[13] = tmp1;
+        diff = tmp0 - state[15];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        state[15] = state[14] + diff * kResampleAllpass[0][2];
+        state[14] = tmp0;
+
+        // average the two allpass outputs, scale down and store
+        out[i << 1] = (out[i << 1] + (state[15] >> 1)) >> 15;
+    }
+}
+
+//   lowpass filter
+// input:  WebRtc_Word32 (shifted 15 positions to the left, + offset 16384)
+// output: WebRtc_Word32 (normalized, not saturated)
+// state:  filter state array; length = 8
+void WebRtcSpl_LPBy2IntToInt(const WebRtc_Word32* in, WebRtc_Word32 len, WebRtc_Word32* out,
+                             WebRtc_Word32* state)
+{
+    WebRtc_Word32 tmp0, tmp1, diff;
+    WebRtc_Word32 i;
+
+    len >>= 1;
+
+    // lower allpass filter: odd input -> even output samples
+    in++;
+    // initial state of polyphase delay element
+    tmp0 = state[12];
+    for (i = 0; i < len; i++)
+    {
+        diff = tmp0 - state[1];
+        // scale down and round
+        diff = (diff + (1 << 13)) >> 14;
+        tmp1 = state[0] + diff * kResampleAllpass[1][0];
+        state[0] = tmp0;
+        diff = tmp1 - state[2];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        tmp0 = state[1] + diff * kResampleAllpass[1][1];
+        state[1] = tmp1;
+        diff = tmp0 - state[3];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        state[3] = state[2] + diff * kResampleAllpass[1][2];
+        state[2] = tmp0;
+
+        // scale down, round and store
+        out[i << 1] = state[3] >> 1;
+        tmp0 = in[i << 1];
+    }
+    in--;
+
+    // upper allpass filter: even input -> even output samples
+    for (i = 0; i < len; i++)
+    {
+        tmp0 = in[i << 1];
+        diff = tmp0 - state[5];
+        // scale down and round
+        diff = (diff + (1 << 13)) >> 14;
+        tmp1 = state[4] + diff * kResampleAllpass[0][0];
+        state[4] = tmp0;
+        diff = tmp1 - state[6];
+        // scale down and round
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        tmp0 = state[5] + diff * kResampleAllpass[0][1];
+        state[5] = tmp1;
+        diff = tmp0 - state[7];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        state[7] = state[6] + diff * kResampleAllpass[0][2];
+        state[6] = tmp0;
+
+        // average the two allpass outputs, scale down and store
+        out[i << 1] = (out[i << 1] + (state[7] >> 1)) >> 15;
+    }
+
+    // switch to odd output samples
+    out++;
+
+    // lower allpass filter: even input -> odd output samples
+    for (i = 0; i < len; i++)
+    {
+        tmp0 = in[i << 1];
+        diff = tmp0 - state[9];
+        // scale down and round
+        diff = (diff + (1 << 13)) >> 14;
+        tmp1 = state[8] + diff * kResampleAllpass[1][0];
+        state[8] = tmp0;
+        diff = tmp1 - state[10];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        tmp0 = state[9] + diff * kResampleAllpass[1][1];
+        state[9] = tmp1;
+        diff = tmp0 - state[11];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        state[11] = state[10] + diff * kResampleAllpass[1][2];
+        state[10] = tmp0;
+
+        // scale down, round and store
+        out[i << 1] = state[11] >> 1;
+    }
+
+    // upper allpass filter: odd input -> odd output samples
+    in++;
+    for (i = 0; i < len; i++)
+    {
+        tmp0 = in[i << 1];
+        diff = tmp0 - state[13];
+        // scale down and round
+        diff = (diff + (1 << 13)) >> 14;
+        tmp1 = state[12] + diff * kResampleAllpass[0][0];
+        state[12] = tmp0;
+        diff = tmp1 - state[14];
+        // scale down and round
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        tmp0 = state[13] + diff * kResampleAllpass[0][1];
+        state[13] = tmp1;
+        diff = tmp0 - state[15];
+        // scale down and truncate
+        diff = diff >> 14;
+        if (diff < 0)
+            diff += 1;
+        state[15] = state[14] + diff * kResampleAllpass[0][2];
+        state[14] = tmp0;
+
+        // average the two allpass outputs, scale down and store
+        out[i << 1] = (out[i << 1] + (state[15] >> 1)) >> 15;
+    }
+}

src/common_audio/signal_processing_library/main/source/resample_by_2_internal.h

@@ -0,0 +1,47 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This header file contains some internal resampling functions.
+ *
+ */
+
+#ifndef WEBRTC_SPL_RESAMPLE_BY_2_INTERNAL_H_
+#define WEBRTC_SPL_RESAMPLE_BY_2_INTERNAL_H_
+
+#include "typedefs.h"
+
+/*******************************************************************
+ * resample_by_2_fast.c
+ * Functions for internal use in the other resample functions
+ ******************************************************************/
+void WebRtcSpl_DownBy2IntToShort(WebRtc_Word32 *in, WebRtc_Word32 len, WebRtc_Word16 *out,
+                                 WebRtc_Word32 *state);
+
+void WebRtcSpl_DownBy2ShortToInt(const WebRtc_Word16 *in, WebRtc_Word32 len,
+                                 WebRtc_Word32 *out, WebRtc_Word32 *state);
+
+void WebRtcSpl_UpBy2ShortToInt(const WebRtc_Word16 *in, WebRtc_Word32 len,
+                               WebRtc_Word32 *out, WebRtc_Word32 *state);
+
+void WebRtcSpl_UpBy2IntToInt(const WebRtc_Word32 *in, WebRtc_Word32 len, WebRtc_Word32 *out,
+                             WebRtc_Word32 *state);
+
+void WebRtcSpl_UpBy2IntToShort(const WebRtc_Word32 *in, WebRtc_Word32 len,
+                               WebRtc_Word16 *out, WebRtc_Word32 *state);
+
+void WebRtcSpl_LPBy2ShortToInt(const WebRtc_Word16* in, WebRtc_Word32 len,
+                               WebRtc_Word32* out, WebRtc_Word32* state);
+
+void WebRtcSpl_LPBy2IntToInt(const WebRtc_Word32* in, WebRtc_Word32 len, WebRtc_Word32* out,
+                             WebRtc_Word32* state);
+
+#endif // WEBRTC_SPL_RESAMPLE_BY_2_INTERNAL_H_

src/common_audio/signal_processing_library/main/source/resample_fractional.c

@@ -0,0 +1,242 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the resampling functions between 48, 44, 32 and 24 kHz.
+ * The description headers can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+// interpolation coefficients
+static const WebRtc_Word16 kCoefficients48To32[2][8] = {
+        {778, -2050, 1087, 23285, 12903, -3783, 441, 222},
+        {222, 441, -3783, 12903, 23285, 1087, -2050, 778}
+};
+
+static const WebRtc_Word16 kCoefficients32To24[3][8] = {
+        {767, -2362, 2434, 24406, 10620, -3838, 721, 90},
+        {386, -381, -2646, 19062, 19062, -2646, -381, 386},
+        {90, 721, -3838, 10620, 24406, 2434, -2362, 767}
+};
+
+static const WebRtc_Word16 kCoefficients44To32[4][9] = {
+        {117, -669, 2245, -6183, 26267, 13529, -3245, 845, -138},
+        {-101, 612, -2283, 8532, 29790, -5138, 1789, -524, 91},
+        {50, -292, 1016, -3064, 32010, 3933, -1147, 315, -53},
+        {-156, 974, -3863, 18603, 21691, -6246, 2353, -712, 126}
+};
+
+//   Resampling ratio: 2/3
+// input:  WebRtc_Word32 (normalized, not saturated) :: size 3 * K
+// output: WebRtc_Word32 (shifted 15 positions to the left, + offset 16384) :: size 2 * K
+//      K: number of blocks
+
+void WebRtcSpl_Resample48khzTo32khz(const WebRtc_Word32 *In, WebRtc_Word32 *Out,
+                                    const WebRtc_Word32 K)
+{
+    /////////////////////////////////////////////////////////////
+    // Filter operation:
+    //
+    // Perform resampling (3 input samples -> 2 output samples);
+    // process in sub blocks of size 3 samples.
+    WebRtc_Word32 tmp;
+    WebRtc_Word32 m;
+
+    for (m = 0; m < K; m++)
+    {
+        tmp = 1 << 14;
+        tmp += kCoefficients48To32[0][0] * In[0];
+        tmp += kCoefficients48To32[0][1] * In[1];
+        tmp += kCoefficients48To32[0][2] * In[2];
+        tmp += kCoefficients48To32[0][3] * In[3];
+        tmp += kCoefficients48To32[0][4] * In[4];
+        tmp += kCoefficients48To32[0][5] * In[5];
+        tmp += kCoefficients48To32[0][6] * In[6];
+        tmp += kCoefficients48To32[0][7] * In[7];
+        Out[0] = tmp;
+
+        tmp = 1 << 14;
+        tmp += kCoefficients48To32[1][0] * In[1];
+        tmp += kCoefficients48To32[1][1] * In[2];
+        tmp += kCoefficients48To32[1][2] * In[3];
+        tmp += kCoefficients48To32[1][3] * In[4];
+        tmp += kCoefficients48To32[1][4] * In[5];
+        tmp += kCoefficients48To32[1][5] * In[6];
+        tmp += kCoefficients48To32[1][6] * In[7];
+        tmp += kCoefficients48To32[1][7] * In[8];
+        Out[1] = tmp;
+
+        // update pointers
+        In += 3;
+        Out += 2;
+    }
+}
+
+//   Resampling ratio: 3/4
+// input:  WebRtc_Word32 (normalized, not saturated) :: size 4 * K
+// output: WebRtc_Word32 (shifted 15 positions to the left, + offset 16384) :: size 3 * K
+//      K: number of blocks
+
+void WebRtcSpl_Resample32khzTo24khz(const WebRtc_Word32 *In, WebRtc_Word32 *Out,
+                                    const WebRtc_Word32 K)
+{
+    /////////////////////////////////////////////////////////////
+    // Filter operation:
+    //
+    // Perform resampling (4 input samples -> 3 output samples);
+    // process in sub blocks of size 4 samples.
+    WebRtc_Word32 m;
+    WebRtc_Word32 tmp;
+
+    for (m = 0; m < K; m++)
+    {
+        tmp = 1 << 14;
+        tmp += kCoefficients32To24[0][0] * In[0];
+        tmp += kCoefficients32To24[0][1] * In[1];
+        tmp += kCoefficients32To24[0][2] * In[2];
+        tmp += kCoefficients32To24[0][3] * In[3];
+        tmp += kCoefficients32To24[0][4] * In[4];
+        tmp += kCoefficients32To24[0][5] * In[5];
+        tmp += kCoefficients32To24[0][6] * In[6];
+        tmp += kCoefficients32To24[0][7] * In[7];
+        Out[0] = tmp;
+
+        tmp = 1 << 14;
+        tmp += kCoefficients32To24[1][0] * In[1];
+        tmp += kCoefficients32To24[1][1] * In[2];
+        tmp += kCoefficients32To24[1][2] * In[3];
+        tmp += kCoefficients32To24[1][3] * In[4];
+        tmp += kCoefficients32To24[1][4] * In[5];
+        tmp += kCoefficients32To24[1][5] * In[6];
+        tmp += kCoefficients32To24[1][6] * In[7];
+        tmp += kCoefficients32To24[1][7] * In[8];
+        Out[1] = tmp;
+
+        tmp = 1 << 14;
+        tmp += kCoefficients32To24[2][0] * In[2];
+        tmp += kCoefficients32To24[2][1] * In[3];
+        tmp += kCoefficients32To24[2][2] * In[4];
+        tmp += kCoefficients32To24[2][3] * In[5];
+        tmp += kCoefficients32To24[2][4] * In[6];
+        tmp += kCoefficients32To24[2][5] * In[7];
+        tmp += kCoefficients32To24[2][6] * In[8];
+        tmp += kCoefficients32To24[2][7] * In[9];
+        Out[2] = tmp;
+
+        // update pointers
+        In += 4;
+        Out += 3;
+    }
+}
+
+//
+// fractional resampling filters
+//   Fout = 11/16 * Fin
+//   Fout =  8/11 * Fin
+//
+
+// compute two inner-products and store them to output array
+static void WebRtcSpl_ResampDotProduct(const WebRtc_Word32 *in1, const WebRtc_Word32 *in2,
+                               const WebRtc_Word16 *coef_ptr, WebRtc_Word32 *out1,
+                               WebRtc_Word32 *out2)
+{
+    WebRtc_Word32 tmp1 = 16384;
+    WebRtc_Word32 tmp2 = 16384;
+    WebRtc_Word16 coef;
+
+    coef = coef_ptr[0];
+    tmp1 += coef * in1[0];
+    tmp2 += coef * in2[-0];
+
+    coef = coef_ptr[1];
+    tmp1 += coef * in1[1];
+    tmp2 += coef * in2[-1];
+
+    coef = coef_ptr[2];
+    tmp1 += coef * in1[2];
+    tmp2 += coef * in2[-2];
+
+    coef = coef_ptr[3];
+    tmp1 += coef * in1[3];
+    tmp2 += coef * in2[-3];
+
+    coef = coef_ptr[4];
+    tmp1 += coef * in1[4];
+    tmp2 += coef * in2[-4];
+
+    coef = coef_ptr[5];
+    tmp1 += coef * in1[5];
+    tmp2 += coef * in2[-5];
+
+    coef = coef_ptr[6];
+    tmp1 += coef * in1[6];
+    tmp2 += coef * in2[-6];
+
+    coef = coef_ptr[7];
+    tmp1 += coef * in1[7];
+    tmp2 += coef * in2[-7];
+
+    coef = coef_ptr[8];
+    *out1 = tmp1 + coef * in1[8];
+    *out2 = tmp2 + coef * in2[-8];
+}
+
+//   Resampling ratio: 8/11
+// input:  WebRtc_Word32 (normalized, not saturated) :: size 11 * K
+// output: WebRtc_Word32 (shifted 15 positions to the left, + offset 16384) :: size  8 * K
+//      K: number of blocks
+
+void WebRtcSpl_Resample44khzTo32khz(const WebRtc_Word32 *In, WebRtc_Word32 *Out,
+                                    const WebRtc_Word32 K)
+{
+    /////////////////////////////////////////////////////////////
+    // Filter operation:
+    //
+    // Perform resampling (11 input samples -> 8 output samples);
+    // process in sub blocks of size 11 samples.
+    WebRtc_Word32 tmp;
+    WebRtc_Word32 m;
+
+    for (m = 0; m < K; m++)
+    {
+        tmp = 1 << 14;
+
+        // first output sample
+        Out[0] = ((WebRtc_Word32)In[3] << 15) + tmp;
+
+        // sum and accumulate filter coefficients and input samples
+        tmp += kCoefficients44To32[3][0] * In[5];
+        tmp += kCoefficients44To32[3][1] * In[6];
+        tmp += kCoefficients44To32[3][2] * In[7];
+        tmp += kCoefficients44To32[3][3] * In[8];
+        tmp += kCoefficients44To32[3][4] * In[9];
+        tmp += kCoefficients44To32[3][5] * In[10];
+        tmp += kCoefficients44To32[3][6] * In[11];
+        tmp += kCoefficients44To32[3][7] * In[12];
+        tmp += kCoefficients44To32[3][8] * In[13];
+        Out[4] = tmp;
+
+        // sum and accumulate filter coefficients and input samples
+        WebRtcSpl_ResampDotProduct(&In[0], &In[17], kCoefficients44To32[0], &Out[1], &Out[7]);
+
+        // sum and accumulate filter coefficients and input samples
+        WebRtcSpl_ResampDotProduct(&In[2], &In[15], kCoefficients44To32[1], &Out[2], &Out[6]);
+
+        // sum and accumulate filter coefficients and input samples
+        WebRtcSpl_ResampDotProduct(&In[3], &In[14], kCoefficients44To32[2], &Out[3], &Out[5]);
+
+        // update pointers
+        In += 11;
+        Out += 8;
+    }
+}

src/common_audio/signal_processing_library/main/source/sin_table.c

@@ -0,0 +1,60 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the 360 degree sine table.
+ *
+ */
+
+#include "signal_processing_library.h"
+
+WebRtc_Word16 WebRtcSpl_kSinTable[] = {
+        0,    142,    285,    428,    571,    713,    856,    998,   1140,
+     1281,   1422,   1563,   1703,   1842,   1981,   2120,   2258,   2395,
+     2531,   2667,   2801,   2935,   3068,   3200,   3331,   3462,   3591,
+     3719,   3845,   3971,   4095,   4219,   4341,   4461,   4580,   4698,
+     4815,   4930,   5043,   5155,   5265,   5374,   5481,   5586,   5690,
+     5792,   5892,   5991,   6087,   6182,   6275,   6366,   6455,   6542,
+     6627,   6710,   6791,   6870,   6947,   7021,   7094,   7164,   7233,
+     7299,   7362,   7424,   7483,   7540,   7595,   7647,   7697,   7745,
+     7791,   7834,   7874,   7912,   7948,   7982,   8012,   8041,   8067,
+     8091,   8112,   8130,   8147,   8160,   8172,   8180,   8187,   8190,
+     8191,   8190,   8187,   8180,   8172,   8160,   8147,   8130,   8112,
+     8091,   8067,   8041,   8012,   7982,   7948,   7912,   7874,   7834,
+     7791,   7745,   7697,   7647,   7595,   7540,   7483,   7424,   7362,
+     7299,   7233,   7164,   7094,   7021,   6947,   6870,   6791,   6710,
+     6627,   6542,   6455,   6366,   6275,   6182,   6087,   5991,   5892,
+     5792,   5690,   5586,   5481,   5374,   5265,   5155,   5043,   4930,
+     4815,   4698,   4580,   4461,   4341,   4219,   4096,   3971,   3845,
+     3719,   3591,   3462,   3331,   3200,   3068,   2935,   2801,   2667,
+     2531,   2395,   2258,   2120,   1981,   1842,   1703,   1563,   1422,
+     1281,   1140,    998,    856,    713,    571,    428,    285,    142,
+        0,   -142,   -285,   -428,   -571,   -713,   -856,   -998,  -1140,
+    -1281,  -1422,  -1563,  -1703,  -1842,  -1981,  -2120,  -2258,  -2395,
+    -2531,  -2667,  -2801,  -2935,  -3068,  -3200,  -3331,  -3462,  -3591,
+    -3719,  -3845,  -3971,  -4095,  -4219,  -4341,  -4461,  -4580,  -4698,
+    -4815,  -4930,  -5043,  -5155,  -5265,  -5374,  -5481,  -5586,  -5690,
+    -5792,  -5892,  -5991,  -6087,  -6182,  -6275,  -6366,  -6455,  -6542,
+    -6627,  -6710,  -6791,  -6870,  -6947,  -7021,  -7094,  -7164,  -7233,
+    -7299,  -7362,  -7424,  -7483,  -7540,  -7595,  -7647,  -7697,  -7745,
+    -7791,  -7834,  -7874,  -7912,  -7948,  -7982,  -8012,  -8041,  -8067,
+    -8091,  -8112,  -8130,  -8147,  -8160,  -8172,  -8180,  -8187,  -8190,
+    -8191,  -8190,  -8187,  -8180,  -8172,  -8160,  -8147,  -8130,  -8112,
+    -8091,  -8067,  -8041,  -8012,  -7982,  -7948,  -7912,  -7874,  -7834,
+    -7791,  -7745,  -7697,  -7647,  -7595,  -7540,  -7483,  -7424,  -7362,
+    -7299,  -7233,  -7164,  -7094,  -7021,  -6947,  -6870,  -6791,  -6710,
+    -6627,  -6542,  -6455,  -6366,  -6275,  -6182,  -6087,  -5991,  -5892,
+    -5792,  -5690,  -5586,  -5481,  -5374,  -5265,  -5155,  -5043,  -4930,
+    -4815,  -4698,  -4580,  -4461,  -4341,  -4219,  -4096,  -3971,  -3845,
+    -3719,  -3591,  -3462,  -3331,  -3200,  -3068,  -2935,  -2801,  -2667,
+    -2531,  -2395,  -2258,  -2120,  -1981,  -1842,  -1703,  -1563,  -1422,
+    -1281,  -1140,   -998,   -856,   -713,   -571,   -428,   -285,   -142
+};

src/common_audio/signal_processing_library/main/source/sin_table_1024.c

@@ -0,0 +1,150 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the 1024 point sine table.
+ *
+ */
+
+#include "signal_processing_library.h"
+
+WebRtc_Word16 WebRtcSpl_kSinTable1024[] =
+{
+      0,    201,    402,    603,    804,   1005,   1206,   1406,
+   1607,   1808,   2009,   2209,   2410,   2610,   2811,   3011,
+   3211,   3411,   3611,   3811,   4011,   4210,   4409,   4608,
+   4807,   5006,   5205,   5403,   5601,   5799,   5997,   6195,
+   6392,   6589,   6786,   6982,   7179,   7375,   7571,   7766,
+   7961,   8156,   8351,   8545,   8739,   8932,   9126,   9319,
+   9511,   9703,   9895,  10087,  10278,  10469,  10659,  10849,
+  11038,  11227,  11416,  11604,  11792,  11980,  12166,  12353,
+  12539,  12724,  12909,  13094,  13278,  13462,  13645,  13827,
+  14009,  14191,  14372,  14552,  14732,  14911,  15090,  15268,
+  15446,  15623,  15799,  15975,  16150,  16325,  16499,  16672,
+  16845,  17017,  17189,  17360,  17530,  17699,  17868,  18036,
+  18204,  18371,  18537,  18702,  18867,  19031,  19194,  19357,
+  19519,  19680,  19840,  20000,  20159,  20317,  20474,  20631,
+  20787,  20942,  21096,  21249,  21402,  21554,  21705,  21855,
+  22004,  22153,  22301,  22448,  22594,  22739,  22883,  23027,
+  23169,  23311,  23452,  23592,  23731,  23869,  24006,  24143,
+  24278,  24413,  24546,  24679,  24811,  24942,  25072,  25201,
+  25329,  25456,  25582,  25707,  25831,  25954,  26077,  26198,
+  26318,  26437,  26556,  26673,  26789,  26905,  27019,  27132,
+  27244,  27355,  27466,  27575,  27683,  27790,  27896,  28001,
+  28105,  28208,  28309,  28410,  28510,  28608,  28706,  28802,
+  28897,  28992,  29085,  29177,  29268,  29358,  29446,  29534,
+  29621,  29706,  29790,  29873,  29955,  30036,  30116,  30195,
+  30272,  30349,  30424,  30498,  30571,  30643,  30713,  30783,
+  30851,  30918,  30984,  31049,
+  31113,  31175,  31236,  31297,
+  31356,  31413,  31470,  31525,  31580,  31633,  31684,  31735,
+  31785,  31833,  31880,  31926,  31970,  32014,  32056,  32097,
+  32137,  32176,  32213,  32249,  32284,  32318,  32350,  32382,
+  32412,  32441,  32468,  32495,  32520,  32544,  32567,  32588,
+  32609,  32628,  32646,  32662,  32678,  32692,  32705,  32717,
+  32727,  32736,  32744,  32751,  32757,  32761,  32764,  32766,
+  32767,  32766,  32764,  32761,  32757,  32751,  32744,  32736,
+  32727,  32717,  32705,  32692,  32678,  32662,  32646,  32628,
+  32609,  32588,  32567,  32544,  32520,  32495,  32468,  32441,
+  32412,  32382,  32350,  32318,  32284,  32249,  32213,  32176,
+  32137,  32097,  32056,  32014,  31970,  31926,  31880,  31833,
+  31785,  31735,  31684,  31633,  31580,  31525,  31470,  31413,
+  31356,  31297,  31236,  31175,  31113,  31049,  30984,  30918,
+  30851,  30783,  30713,  30643,  30571,  30498,  30424,  30349,
+  30272,  30195,  30116,  30036,  29955,  29873,  29790,  29706,
+  29621,  29534,  29446,  29358,  29268,  29177,  29085,  28992,
+  28897,  28802,  28706,  28608,  28510,  28410,  28309,  28208,
+  28105,  28001,  27896,  27790,  27683,  27575,  27466,  27355,
+  27244,  27132,  27019,  26905,  26789,  26673,  26556,  26437,
+  26318,  26198,  26077,  25954,  25831,  25707,  25582,  25456,
+  25329,  25201,  25072,  24942,  24811,  24679,  24546,  24413,
+  24278,  24143,  24006,  23869,  23731,  23592,  23452,  23311,
+  23169,  23027,  22883,  22739,  22594,  22448,  22301,  22153,
+  22004,  21855,  21705,  21554,  21402,  21249,  21096,  20942,
+  20787,  20631,  20474,  20317,  20159,  20000,  19840,  19680,
+  19519,  19357,  19194,  19031,  18867,  18702,  18537,  18371,
+  18204,  18036,  17868,  17699,  17530,  17360,  17189,  17017,
+  16845,  16672,  16499,  16325,  16150,  15975,  15799,  15623,
+  15446,  15268,  15090,  14911,  14732,  14552,  14372,  14191,
+  14009,  13827,  13645,  13462,  13278,  13094,  12909,  12724,
+  12539,  12353,  12166,  11980,  11792,  11604,  11416,  11227,
+  11038,  10849,  10659,  10469,  10278,  10087,   9895,   9703,
+   9511,   9319,   9126,   8932,   8739,   8545,   8351,   8156,
+   7961,   7766,   7571,   7375,   7179,   6982,   6786,   6589,
+   6392,   6195,   5997,   5799,   5601,   5403,   5205,   5006,
+   4807,   4608,   4409,   4210,   4011,   3811,   3611,   3411,
+   3211,   3011,   2811,   2610,   2410,   2209,   2009,   1808,
+   1607,   1406,   1206,   1005,    804,    603,    402,    201,
+      0,   -201,   -402,   -603,   -804,  -1005,  -1206,  -1406,
+  -1607,  -1808,  -2009,  -2209,  -2410,  -2610,  -2811,  -3011,
+  -3211,  -3411,  -3611,  -3811,  -4011,  -4210,  -4409,  -4608,
+  -4807,  -5006,  -5205,  -5403,  -5601,  -5799,  -5997,  -6195,
+  -6392,  -6589,  -6786,  -6982,  -7179,  -7375,  -7571,  -7766,
+  -7961,  -8156,  -8351,  -8545,  -8739,  -8932,  -9126,  -9319,
+  -9511,  -9703,  -9895, -10087, -10278, -10469, -10659, -10849,
+ -11038, -11227, -11416, -11604, -11792, -11980, -12166, -12353,
+ -12539, -12724, -12909, -13094, -13278, -13462, -13645, -13827,
+ -14009, -14191, -14372, -14552, -14732, -14911, -15090, -15268,
+ -15446, -15623, -15799, -15975, -16150, -16325, -16499, -16672,
+ -16845, -17017, -17189, -17360, -17530, -17699, -17868, -18036,
+ -18204, -18371, -18537, -18702, -18867, -19031, -19194, -19357,
+ -19519, -19680, -19840, -20000, -20159, -20317, -20474, -20631,
+ -20787, -20942, -21096, -21249, -21402, -21554, -21705, -21855,
+ -22004, -22153, -22301, -22448, -22594, -22739, -22883, -23027,
+ -23169, -23311, -23452, -23592, -23731, -23869, -24006, -24143,
+ -24278, -24413, -24546, -24679, -24811, -24942, -25072, -25201,
+ -25329, -25456, -25582, -25707, -25831, -25954, -26077, -26198,
+ -26318, -26437, -26556, -26673, -26789, -26905, -27019, -27132,
+ -27244, -27355, -27466, -27575, -27683, -27790, -27896, -28001,
+ -28105, -28208, -28309, -28410, -28510, -28608, -28706, -28802,
+ -28897, -28992, -29085, -29177, -29268, -29358, -29446, -29534,
+ -29621, -29706, -29790, -29873, -29955, -30036, -30116, -30195,
+ -30272, -30349, -30424, -30498, -30571, -30643, -30713, -30783,
+ -30851, -30918, -30984, -31049, -31113, -31175, -31236, -31297,
+ -31356, -31413, -31470, -31525, -31580, -31633, -31684, -31735,
+ -31785, -31833, -31880, -31926, -31970, -32014, -32056, -32097,
+ -32137, -32176, -32213, -32249, -32284, -32318, -32350, -32382,
+ -32412, -32441, -32468, -32495, -32520, -32544, -32567, -32588,
+ -32609, -32628, -32646, -32662, -32678, -32692, -32705, -32717,
+ -32727, -32736, -32744, -32751, -32757, -32761, -32764, -32766,
+ -32767, -32766, -32764, -32761, -32757, -32751, -32744, -32736,
+ -32727, -32717, -32705, -32692, -32678, -32662, -32646, -32628,
+ -32609, -32588, -32567, -32544, -32520, -32495, -32468, -32441,
+ -32412, -32382, -32350, -32318, -32284, -32249, -32213, -32176,
+ -32137, -32097, -32056, -32014, -31970, -31926, -31880, -31833,
+ -31785, -31735, -31684, -31633, -31580, -31525, -31470, -31413,
+ -31356, -31297, -31236, -31175, -31113, -31049, -30984, -30918,
+ -30851, -30783, -30713, -30643, -30571, -30498, -30424, -30349,
+ -30272, -30195, -30116, -30036, -29955, -29873, -29790, -29706,
+ -29621, -29534, -29446, -29358, -29268, -29177, -29085, -28992,
+ -28897, -28802, -28706, -28608, -28510, -28410, -28309, -28208,
+ -28105, -28001, -27896, -27790, -27683, -27575, -27466, -27355,
+ -27244, -27132, -27019, -26905, -26789, -26673, -26556, -26437,
+ -26318, -26198, -26077, -25954, -25831, -25707, -25582, -25456,
+ -25329, -25201, -25072, -24942, -24811, -24679, -24546, -24413,
+ -24278, -24143, -24006, -23869, -23731, -23592, -23452, -23311,
+ -23169, -23027, -22883, -22739, -22594, -22448, -22301, -22153,
+ -22004, -21855, -21705, -21554, -21402, -21249, -21096, -20942,
+ -20787, -20631, -20474, -20317, -20159, -20000, -19840, -19680,
+ -19519, -19357, -19194, -19031, -18867, -18702, -18537, -18371,
+ -18204, -18036, -17868, -17699, -17530, -17360, -17189, -17017,
+ -16845, -16672, -16499, -16325, -16150, -15975, -15799, -15623,
+ -15446, -15268, -15090, -14911, -14732, -14552, -14372, -14191,
+ -14009, -13827, -13645, -13462, -13278, -13094, -12909, -12724,
+ -12539, -12353, -12166, -11980, -11792, -11604, -11416, -11227,
+ -11038, -10849, -10659, -10469, -10278, -10087,  -9895,  -9703,
+  -9511,  -9319,  -9126,  -8932,  -8739,  -8545,  -8351,  -8156,
+  -7961,  -7766,  -7571,  -7375,  -7179,  -6982,  -6786,  -6589,
+  -6392,  -6195,  -5997,  -5799,  -5601,  -5403,  -5205,  -5006,
+  -4807,  -4608,  -4409,  -4210,  -4011,  -3811,  -3611,  -3411,
+  -3211,  -3011,  -2811,  -2610,  -2410,  -2209,  -2009,  -1808,
+  -1607,  -1406,  -1206,  -1005,   -804,   -603,   -402,   -201,
+};

src/common_audio/signal_processing_library/main/source/spl.gypi

@@ -0,0 +1,73 @@
+# Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+#
+# Use of this source code is governed by a BSD-style license
+# that can be found in the LICENSE file in the root of the source
+# tree. An additional intellectual property rights grant can be found
+# in the file PATENTS.  All contributing project authors may
+# be found in the AUTHORS file in the root of the source tree.
+
+{
+  'targets': [
+    {
+      'target_name': 'spl',
+      'type': '<(library)',
+      'include_dirs': [
+        '../interface',
+      ],
+      'direct_dependent_settings': {
+        'include_dirs': [
+          '../interface',
+        ],
+      },
+      'sources': [
+        '../interface/signal_processing_library.h',
+        '../interface/spl_inl.h',
+        'auto_corr_to_refl_coef.c',
+        'auto_correlation.c',
+        'complex_fft.c',
+        'complex_ifft.c',
+        'complex_bit_reverse.c',
+        'copy_set_operations.c',
+        'cos_table.c',
+        'cross_correlation.c',
+        'division_operations.c',
+        'dot_product_with_scale.c',
+        'downsample_fast.c',
+        'energy.c',
+        'filter_ar.c',
+        'filter_ar_fast_q12.c',
+        'filter_ma_fast_q12.c',
+        'get_hanning_window.c',
+        'get_scaling_square.c',
+        'hanning_table.c',
+        'ilbc_specific_functions.c',
+        'levinson_durbin.c',
+        'lpc_to_refl_coef.c',
+        'min_max_operations.c',
+        'randn_table.c',
+        'randomization_functions.c',
+        'refl_coef_to_lpc.c',
+        'resample.c',
+        'resample_48khz.c',
+        'resample_by_2.c',
+        'resample_by_2_internal.c',
+        'resample_by_2_internal.h',
+        'resample_fractional.c',
+        'sin_table.c',
+        'sin_table_1024.c',
+        'spl_sqrt.c',
+        'spl_sqrt_floor.c',
+        'spl_version.c',
+        'splitting_filter.c',
+        'sqrt_of_one_minus_x_squared.c',
+        'vector_scaling_operations.c',
+      ],
+    },
+  ],
+}
+
+# Local Variables:
+# tab-width:2
+# indent-tabs-mode:nil
+# End:
+# vim: set expandtab tabstop=2 shiftwidth=2:

src/common_audio/signal_processing_library/main/source/spl_sqrt.c

@@ -0,0 +1,184 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_Sqrt().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+WebRtc_Word32 WebRtcSpl_SqrtLocal(WebRtc_Word32 in);
+
+WebRtc_Word32 WebRtcSpl_SqrtLocal(WebRtc_Word32 in)
+{
+
+    WebRtc_Word16 x_half, t16;
+    WebRtc_Word32 A, B, x2;
+
+    /* The following block performs:
+     y=in/2
+     x=y-2^30
+     x_half=x/2^31
+     t = 1 + (x_half) - 0.5*((x_half)^2) + 0.5*((x_half)^3) - 0.625*((x_half)^4)
+         + 0.875*((x_half)^5)
+     */
+
+    B = in;
+
+    B = WEBRTC_SPL_RSHIFT_W32(B, 1); // B = in/2
+    B = B - ((WebRtc_Word32)0x40000000); // B = in/2 - 1/2
+    x_half = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(B, 16);// x_half = x/2 = (in-1)/2
+    B = B + ((WebRtc_Word32)0x40000000); // B = 1 + x/2
+    B = B + ((WebRtc_Word32)0x40000000); // Add 0.5 twice (since 1.0 does not exist in Q31)
+
+    x2 = ((WebRtc_Word32)x_half) * ((WebRtc_Word32)x_half) * 2; // A = (x/2)^2
+    A = -x2; // A = -(x/2)^2
+    B = B + (A >> 1); // B = 1 + x/2 - 0.5*(x/2)^2
+
+    A = WEBRTC_SPL_RSHIFT_W32(A, 16);
+    A = A * A * 2; // A = (x/2)^4
+    t16 = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(A, 16);
+    B = B + WEBRTC_SPL_MUL_16_16(-20480, t16) * 2; // B = B - 0.625*A
+    // After this, B = 1 + x/2 - 0.5*(x/2)^2 - 0.625*(x/2)^4
+
+    t16 = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(A, 16);
+    A = WEBRTC_SPL_MUL_16_16(x_half, t16) * 2; // A = (x/2)^5
+    t16 = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(A, 16);
+    B = B + WEBRTC_SPL_MUL_16_16(28672, t16) * 2; // B = B + 0.875*A
+    // After this, B = 1 + x/2 - 0.5*(x/2)^2 - 0.625*(x/2)^4 + 0.875*(x/2)^5
+
+    t16 = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(x2, 16);
+    A = WEBRTC_SPL_MUL_16_16(x_half, t16) * 2; // A = x/2^3
+
+    B = B + (A >> 1); // B = B + 0.5*A
+    // After this, B = 1 + x/2 - 0.5*(x/2)^2 + 0.5*(x/2)^3 - 0.625*(x/2)^4 + 0.875*(x/2)^5
+
+    B = B + ((WebRtc_Word32)32768); // Round off bit
+
+    return B;
+}
+
+WebRtc_Word32 WebRtcSpl_Sqrt(WebRtc_Word32 value)
+{
+    /*
+     Algorithm:
+
+     Six term Taylor Series is used here to compute the square root of a number
+     y^0.5 = (1+x)^0.5 where x = y-1
+     = 1+(x/2)-0.5*((x/2)^2+0.5*((x/2)^3-0.625*((x/2)^4+0.875*((x/2)^5)
+     0.5 <= x < 1
+
+     Example of how the algorithm works, with ut=sqrt(in), and
+     with in=73632 and ut=271 (even shift value case):
+
+     in=73632
+     y= in/131072
+     x=y-1
+     t = 1 + (x/2) - 0.5*((x/2)^2) + 0.5*((x/2)^3) - 0.625*((x/2)^4) + 0.875*((x/2)^5)
+     ut=t*(1/sqrt(2))*512
+
+     or:
+
+     in=73632
+     in2=73632*2^14
+     y= in2/2^31
+     x=y-1
+     t = 1 + (x/2) - 0.5*((x/2)^2) + 0.5*((x/2)^3) - 0.625*((x/2)^4) + 0.875*((x/2)^5)
+     ut=t*(1/sqrt(2))
+     ut2=ut*2^9
+
+     which gives:
+
+     in  = 73632
+     in2 = 1206386688
+     y   = 0.56176757812500
+     x   = -0.43823242187500
+     t   = 0.74973506527313
+     ut  = 0.53014274874797
+     ut2 = 2.714330873589594e+002
+
+     or:
+
+     in=73632
+     in2=73632*2^14
+     y=in2/2
+     x=y-2^30
+     x_half=x/2^31
+     t = 1 + (x_half) - 0.5*((x_half)^2) + 0.5*((x_half)^3) - 0.625*((x_half)^4)
+         + 0.875*((x_half)^5)
+     ut=t*(1/sqrt(2))
+     ut2=ut*2^9
+
+     which gives:
+
+     in  = 73632
+     in2 = 1206386688
+     y   = 603193344
+     x   = -470548480
+     x_half =  -0.21911621093750
+     t   = 0.74973506527313
+     ut  = 0.53014274874797
+     ut2 = 2.714330873589594e+002
+
+     */
+
+    WebRtc_Word16 x_norm, nshift, t16, sh;
+    WebRtc_Word32 A;
+
+    WebRtc_Word16 k_sqrt_2 = 23170; // 1/sqrt2 (==5a82)
+
+    A = value;
+
+    if (A == 0)
+        return (WebRtc_Word32)0; // sqrt(0) = 0
+
+    sh = WebRtcSpl_NormW32(A); // # shifts to normalize A
+    A = WEBRTC_SPL_LSHIFT_W32(A, sh); // Normalize A
+    if (A < (WEBRTC_SPL_WORD32_MAX - 32767))
+    {
+        A = A + ((WebRtc_Word32)32768); // Round off bit
+    } else
+    {
+        A = WEBRTC_SPL_WORD32_MAX;
+    }
+
+    x_norm = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(A, 16); // x_norm = AH
+
+    nshift = WEBRTC_SPL_RSHIFT_W16(sh, 1); // nshift = sh>>1
+    nshift = -nshift; // Negate the power for later de-normalization
+
+    A = (WebRtc_Word32)WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)x_norm, 16);
+    A = WEBRTC_SPL_ABS_W32(A); // A = abs(x_norm<<16)
+    A = WebRtcSpl_SqrtLocal(A); // A = sqrt(A)
+
+    if ((-2 * nshift) == sh)
+    { // Even shift value case
+
+        t16 = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(A, 16); // t16 = AH
+
+        A = WEBRTC_SPL_MUL_16_16(k_sqrt_2, t16) * 2; // A = 1/sqrt(2)*t16
+        A = A + ((WebRtc_Word32)32768); // Round off
+        A = A & ((WebRtc_Word32)0x7fff0000); // Round off
+
+        A = WEBRTC_SPL_RSHIFT_W32(A, 15); // A = A>>16
+
+    } else
+    {
+        A = WEBRTC_SPL_RSHIFT_W32(A, 16); // A = A>>16
+    }
+
+    A = A & ((WebRtc_Word32)0x0000ffff);
+    A = (WebRtc_Word32)WEBRTC_SPL_SHIFT_W32(A, nshift); // De-normalize the result
+
+    return A;
+}

src/common_audio/signal_processing_library/main/source/spl_sqrt_floor.c

@@ -0,0 +1,53 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+/*
+ * This file contains the function WebRtcSpl_SqrtFloor().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+#define WEBRTC_SPL_SQRT_ITER(N)                 \
+  try1 = root + (1 << (N));                     \
+  if (value >= try1 << (N))                     \
+  {                                             \
+    value -= try1 << (N);                       \
+    root |= 2 << (N);                           \
+  }
+
+// (out) Square root of input parameter
+WebRtc_Word32 WebRtcSpl_SqrtFloor(WebRtc_Word32 value)
+{
+    // new routine for performance, 4 cycles/bit in ARM
+    // output precision is 16 bits
+
+    WebRtc_Word32 root = 0, try1;
+
+    WEBRTC_SPL_SQRT_ITER (15);
+    WEBRTC_SPL_SQRT_ITER (14);
+    WEBRTC_SPL_SQRT_ITER (13);
+    WEBRTC_SPL_SQRT_ITER (12);
+    WEBRTC_SPL_SQRT_ITER (11);
+    WEBRTC_SPL_SQRT_ITER (10);
+    WEBRTC_SPL_SQRT_ITER ( 9);
+    WEBRTC_SPL_SQRT_ITER ( 8);
+    WEBRTC_SPL_SQRT_ITER ( 7);
+    WEBRTC_SPL_SQRT_ITER ( 6);
+    WEBRTC_SPL_SQRT_ITER ( 5);
+    WEBRTC_SPL_SQRT_ITER ( 4);
+    WEBRTC_SPL_SQRT_ITER ( 3);
+    WEBRTC_SPL_SQRT_ITER ( 2);
+    WEBRTC_SPL_SQRT_ITER ( 1);
+    WEBRTC_SPL_SQRT_ITER ( 0);
+
+    return root >> 1;
+}

src/common_audio/signal_processing_library/main/source/spl_version.c

@@ -0,0 +1,25 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_get_version().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include <string.h>
+#include "signal_processing_library.h"
+
+WebRtc_Word16 WebRtcSpl_get_version(char* version, WebRtc_Word16 length_in_bytes)
+{
+    strncpy(version, "1.2.0", length_in_bytes);
+    return 0;
+}

src/common_audio/signal_processing_library/main/source/splitting_filter.c

@@ -0,0 +1,200 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+/*
+ * This file contains the splitting filter functions.
+ *
+ */
+
+#include "signal_processing_library.h"
+
+// Number of samples in a low/high-band frame.
+enum
+{
+    kBandFrameLength = 160
+};
+
+// QMF filter coefficients in Q16.
+static const WebRtc_UWord16 WebRtcSpl_kAllPassFilter1[3] = {6418, 36982, 57261};
+static const WebRtc_UWord16 WebRtcSpl_kAllPassFilter2[3] = {21333, 49062, 63010};
+
+///////////////////////////////////////////////////////////////////////////////////////////////
+// WebRtcSpl_AllPassQMF(...)
+//
+// Allpass filter used by the analysis and synthesis parts of the QMF filter.
+//
+// Input:
+//    - in_data             : Input data sequence (Q10)
+//    - data_length         : Length of data sequence (>2)
+//    - filter_coefficients : Filter coefficients (length 3, Q16)
+//
+// Input & Output:
+//    - filter_state        : Filter state (length 6, Q10).
+//
+// Output:
+//    - out_data            : Output data sequence (Q10), length equal to
+//                            |data_length|
+//
+
+void WebRtcSpl_AllPassQMF(WebRtc_Word32* in_data, const WebRtc_Word16 data_length,
+                          WebRtc_Word32* out_data, const WebRtc_UWord16* filter_coefficients,
+                          WebRtc_Word32* filter_state)
+{
+    // The procedure is to filter the input with three first order all pass filters
+    // (cascade operations).
+    //
+    //         a_3 + q^-1    a_2 + q^-1    a_1 + q^-1
+    // y[n] =  -----------   -----------   -----------   x[n]
+    //         1 + a_3q^-1   1 + a_2q^-1   1 + a_1q^-1
+    //
+    // The input vector |filter_coefficients| includes these three filter coefficients.
+    // The filter state contains the in_data state, in_data[-1], followed by
+    // the out_data state, out_data[-1]. This is repeated for each cascade.
+    // The first cascade filter will filter the |in_data| and store the output in
+    // |out_data|. The second will the take the |out_data| as input and make an
+    // intermediate storage in |in_data|, to save memory. The third, and final, cascade
+    // filter operation takes the |in_data| (which is the output from the previous cascade
+    // filter) and store the output in |out_data|.
+    // Note that the input vector values are changed during the process.
+    WebRtc_Word16 k;
+    WebRtc_Word32 diff;
+    // First all-pass cascade; filter from in_data to out_data.
+
+    // Let y_i[n] indicate the output of cascade filter i (with filter coefficient a_i) at
+    // vector position n. Then the final output will be y[n] = y_3[n]
+
+    // First loop, use the states stored in memory.
+    // "diff" should be safe from wrap around since max values are 2^25
+    diff = WEBRTC_SPL_SUB_SAT_W32(in_data[0], filter_state[1]); // = (x[0] - y_1[-1])
+    // y_1[0] =  x[-1] + a_1 * (x[0] - y_1[-1])
+    out_data[0] = WEBRTC_SPL_SCALEDIFF32(filter_coefficients[0], diff, filter_state[0]);
+
+    // For the remaining loops, use previous values.
+    for (k = 1; k < data_length; k++)
+    {
+        diff = WEBRTC_SPL_SUB_SAT_W32(in_data[k], out_data[k - 1]); // = (x[n] - y_1[n-1])
+        // y_1[n] =  x[n-1] + a_1 * (x[n] - y_1[n-1])
+        out_data[k] = WEBRTC_SPL_SCALEDIFF32(filter_coefficients[0], diff, in_data[k - 1]);
+    }
+
+    // Update states.
+    filter_state[0] = in_data[data_length - 1]; // x[N-1], becomes x[-1] next time
+    filter_state[1] = out_data[data_length - 1]; // y_1[N-1], becomes y_1[-1] next time
+
+    // Second all-pass cascade; filter from out_data to in_data.
+    diff = WEBRTC_SPL_SUB_SAT_W32(out_data[0], filter_state[3]); // = (y_1[0] - y_2[-1])
+    // y_2[0] =  y_1[-1] + a_2 * (y_1[0] - y_2[-1])
+    in_data[0] = WEBRTC_SPL_SCALEDIFF32(filter_coefficients[1], diff, filter_state[2]);
+    for (k = 1; k < data_length; k++)
+    {
+        diff = WEBRTC_SPL_SUB_SAT_W32(out_data[k], in_data[k - 1]); // =(y_1[n] - y_2[n-1])
+        // y_2[0] =  y_1[-1] + a_2 * (y_1[0] - y_2[-1])
+        in_data[k] = WEBRTC_SPL_SCALEDIFF32(filter_coefficients[1], diff, out_data[k-1]);
+    }
+
+    filter_state[2] = out_data[data_length - 1]; // y_1[N-1], becomes y_1[-1] next time
+    filter_state[3] = in_data[data_length - 1]; // y_2[N-1], becomes y_2[-1] next time
+
+    // Third all-pass cascade; filter from in_data to out_data.
+    diff = WEBRTC_SPL_SUB_SAT_W32(in_data[0], filter_state[5]); // = (y_2[0] - y[-1])
+    // y[0] =  y_2[-1] + a_3 * (y_2[0] - y[-1])
+    out_data[0] = WEBRTC_SPL_SCALEDIFF32(filter_coefficients[2], diff, filter_state[4]);
+    for (k = 1; k < data_length; k++)
+    {
+        diff = WEBRTC_SPL_SUB_SAT_W32(in_data[k], out_data[k - 1]); // = (y_2[n] - y[n-1])
+        // y[n] =  y_2[n-1] + a_3 * (y_2[n] - y[n-1])
+        out_data[k] = WEBRTC_SPL_SCALEDIFF32(filter_coefficients[2], diff, in_data[k-1]);
+    }
+    filter_state[4] = in_data[data_length - 1]; // y_2[N-1], becomes y_2[-1] next time
+    filter_state[5] = out_data[data_length - 1]; // y[N-1], becomes y[-1] next time
+}
+
+void WebRtcSpl_AnalysisQMF(const WebRtc_Word16* in_data, WebRtc_Word16* low_band,
+                           WebRtc_Word16* high_band, WebRtc_Word32* filter_state1,
+                           WebRtc_Word32* filter_state2)
+{
+    WebRtc_Word16 i;
+    WebRtc_Word16 k;
+    WebRtc_Word32 tmp;
+    WebRtc_Word32 half_in1[kBandFrameLength];
+    WebRtc_Word32 half_in2[kBandFrameLength];
+    WebRtc_Word32 filter1[kBandFrameLength];
+    WebRtc_Word32 filter2[kBandFrameLength];
+
+    // Split even and odd samples. Also shift them to Q10.
+    for (i = 0, k = 0; i < kBandFrameLength; i++, k += 2)
+    {
+        half_in2[i] = WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)in_data[k], 10);
+        half_in1[i] = WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)in_data[k + 1], 10);
+    }
+
+    // All pass filter even and odd samples, independently.
+    WebRtcSpl_AllPassQMF(half_in1, kBandFrameLength, filter1, WebRtcSpl_kAllPassFilter1,
+                         filter_state1);
+    WebRtcSpl_AllPassQMF(half_in2, kBandFrameLength, filter2, WebRtcSpl_kAllPassFilter2,
+                         filter_state2);
+
+    // Take the sum and difference of filtered version of odd and even
+    // branches to get upper & lower band.
+    for (i = 0; i < kBandFrameLength; i++)
+    {
+        tmp = filter1[i] + filter2[i] + 1024;
+        tmp = WEBRTC_SPL_RSHIFT_W32(tmp, 11);
+        low_band[i] = (WebRtc_Word16)WEBRTC_SPL_SAT(WEBRTC_SPL_WORD16_MAX,
+                tmp, WEBRTC_SPL_WORD16_MIN);
+
+        tmp = filter1[i] - filter2[i] + 1024;
+        tmp = WEBRTC_SPL_RSHIFT_W32(tmp, 11);
+        high_band[i] = (WebRtc_Word16)WEBRTC_SPL_SAT(WEBRTC_SPL_WORD16_MAX,
+                tmp, WEBRTC_SPL_WORD16_MIN);
+    }
+}
+
+void WebRtcSpl_SynthesisQMF(const WebRtc_Word16* low_band, const WebRtc_Word16* high_band,
+                            WebRtc_Word16* out_data, WebRtc_Word32* filter_state1,
+                            WebRtc_Word32* filter_state2)
+{
+    WebRtc_Word32 tmp;
+    WebRtc_Word32 half_in1[kBandFrameLength];
+    WebRtc_Word32 half_in2[kBandFrameLength];
+    WebRtc_Word32 filter1[kBandFrameLength];
+    WebRtc_Word32 filter2[kBandFrameLength];
+    WebRtc_Word16 i;
+    WebRtc_Word16 k;
+
+    // Obtain the sum and difference channels out of upper and lower-band channels.
+    // Also shift to Q10 domain.
+    for (i = 0; i < kBandFrameLength; i++)
+    {
+        tmp = (WebRtc_Word32)low_band[i] + (WebRtc_Word32)high_band[i];
+        half_in1[i] = WEBRTC_SPL_LSHIFT_W32(tmp, 10);
+        tmp = (WebRtc_Word32)low_band[i] - (WebRtc_Word32)high_band[i];
+        half_in2[i] = WEBRTC_SPL_LSHIFT_W32(tmp, 10);
+    }
+
+    // all-pass filter the sum and difference channels
+    WebRtcSpl_AllPassQMF(half_in1, kBandFrameLength, filter1, WebRtcSpl_kAllPassFilter2,
+                         filter_state1);
+    WebRtcSpl_AllPassQMF(half_in2, kBandFrameLength, filter2, WebRtcSpl_kAllPassFilter1,
+                         filter_state2);
+
+    // The filtered signals are even and odd samples of the output. Combine
+    // them. The signals are Q10 should shift them back to Q0 and take care of
+    // saturation.
+    for (i = 0, k = 0; i < kBandFrameLength; i++)
+    {
+        tmp = WEBRTC_SPL_RSHIFT_W32(filter2[i] + 512, 10);
+        out_data[k++] = (WebRtc_Word16)WEBRTC_SPL_SAT(32767, tmp, -32768);
+
+        tmp = WEBRTC_SPL_RSHIFT_W32(filter1[i] + 512, 10);
+        out_data[k++] = (WebRtc_Word16)WEBRTC_SPL_SAT(32767, tmp, -32768);
+    }
+
+}

src/common_audio/signal_processing_library/main/source/sqrt_of_one_minus_x_squared.c

@@ -0,0 +1,35 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the function WebRtcSpl_SqrtOfOneMinusXSquared().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+void WebRtcSpl_SqrtOfOneMinusXSquared(WebRtc_Word16 *xQ15, int vector_length,
+                                      WebRtc_Word16 *yQ15)
+{
+    WebRtc_Word32 sq;
+    int m;
+    WebRtc_Word16 tmp;
+
+    for (m = 0; m < vector_length; m++)
+    {
+        tmp = xQ15[m];
+        sq = WEBRTC_SPL_MUL_16_16(tmp, tmp); // x^2 in Q30
+        sq = 1073741823 - sq; // 1-x^2, where 1 ~= 0.99999999906 is 1073741823 in Q30
+        sq = WebRtcSpl_Sqrt(sq); // sqrt(1-x^2) in Q15
+        yQ15[m] = (WebRtc_Word16)sq;
+    }
+}

src/common_audio/signal_processing_library/main/source/vector_scaling_operations.c

@@ -0,0 +1,151 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains implementations of the functions
+ * WebRtcSpl_VectorBitShiftW16()
+ * WebRtcSpl_VectorBitShiftW32()
+ * WebRtcSpl_VectorBitShiftW32ToW16()
+ * WebRtcSpl_ScaleVector()
+ * WebRtcSpl_ScaleVectorWithSat()
+ * WebRtcSpl_ScaleAndAddVectors()
+ *
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#include "signal_processing_library.h"
+
+void WebRtcSpl_VectorBitShiftW16(WebRtc_Word16 *res,
+                             WebRtc_Word16 length,
+                             G_CONST WebRtc_Word16 *in,
+                             WebRtc_Word16 right_shifts)
+{
+    int i;
+
+    if (right_shifts > 0)
+    {
+        for (i = length; i > 0; i--)
+        {
+            (*res++) = ((*in++) >> right_shifts);
+        }
+    } else
+    {
+        for (i = length; i > 0; i--)
+        {
+            (*res++) = ((*in++) << (-right_shifts));
+        }
+    }
+}
+
+void WebRtcSpl_VectorBitShiftW32(WebRtc_Word32 *out_vector,
+                             WebRtc_Word16 vector_length,
+                             G_CONST WebRtc_Word32 *in_vector,
+                             WebRtc_Word16 right_shifts)
+{
+    int i;
+
+    if (right_shifts > 0)
+    {
+        for (i = vector_length; i > 0; i--)
+        {
+            (*out_vector++) = ((*in_vector++) >> right_shifts);
+        }
+    } else
+    {
+        for (i = vector_length; i > 0; i--)
+        {
+            (*out_vector++) = ((*in_vector++) << (-right_shifts));
+        }
+    }
+}
+
+void WebRtcSpl_VectorBitShiftW32ToW16(WebRtc_Word16 *res,
+                                  WebRtc_Word16 length,
+                                  G_CONST WebRtc_Word32 *in,
+                                  WebRtc_Word16 right_shifts)
+{
+    int i;
+
+    if (right_shifts >= 0)
+    {
+        for (i = length; i > 0; i--)
+        {
+            (*res++) = (WebRtc_Word16)((*in++) >> right_shifts);
+        }
+    } else
+    {
+        WebRtc_Word16 left_shifts = -right_shifts;
+        for (i = length; i > 0; i--)
+        {
+            (*res++) = (WebRtc_Word16)((*in++) << left_shifts);
+        }
+    }
+}
+
+void WebRtcSpl_ScaleVector(G_CONST WebRtc_Word16 *in_vector, WebRtc_Word16 *out_vector,
+                           WebRtc_Word16 gain, WebRtc_Word16 in_vector_length,
+                           WebRtc_Word16 right_shifts)
+{
+    // Performs vector operation: out_vector = (gain*in_vector)>>right_shifts
+    int i;
+    G_CONST WebRtc_Word16 *inptr;
+    WebRtc_Word16 *outptr;
+
+    inptr = in_vector;
+    outptr = out_vector;
+
+    for (i = 0; i < in_vector_length; i++)
+    {
+        (*outptr++) = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(*inptr++, gain, right_shifts);
+    }
+}
+
+void WebRtcSpl_ScaleVectorWithSat(G_CONST WebRtc_Word16 *in_vector, WebRtc_Word16 *out_vector,
+                                 WebRtc_Word16 gain, WebRtc_Word16 in_vector_length,
+                                 WebRtc_Word16 right_shifts)
+{
+    // Performs vector operation: out_vector = (gain*in_vector)>>right_shifts
+    int i;
+    WebRtc_Word32 tmpW32;
+    G_CONST WebRtc_Word16 *inptr;
+    WebRtc_Word16 *outptr;
+
+    inptr = in_vector;
+    outptr = out_vector;
+
+    for (i = 0; i < in_vector_length; i++)
+    {
+        tmpW32 = WEBRTC_SPL_MUL_16_16_RSFT(*inptr++, gain, right_shifts);
+        ( *outptr++) = (WebRtc_Word16)WEBRTC_SPL_SAT(32767, tmpW32, -32768);
+    }
+}
+
+void WebRtcSpl_ScaleAndAddVectors(G_CONST WebRtc_Word16 *in1, WebRtc_Word16 gain1, int shift1,
+                                  G_CONST WebRtc_Word16 *in2, WebRtc_Word16 gain2, int shift2,
+                                  WebRtc_Word16 *out, int vector_length)
+{
+    // Performs vector operation: out = (gain1*in1)>>shift1 + (gain2*in2)>>shift2
+    int i;
+    G_CONST WebRtc_Word16 *in1ptr;
+    G_CONST WebRtc_Word16 *in2ptr;
+    WebRtc_Word16 *outptr;
+
+    in1ptr = in1;
+    in2ptr = in2;
+    outptr = out;
+
+    for (i = 0; i < vector_length; i++)
+    {
+        (*outptr++) = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(gain1, *in1ptr++, shift1)
+                + (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(gain2, *in2ptr++, shift2);
+    }
+}

src/common_audio/signal_processing_library/main/source/webrtc_fft_t_1024_8.c

@@ -0,0 +1,704 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the Q14 radix-8 tables used in ARM9e optimizations.
+ *
+ */
+
+extern const int s_Q14S_8;
+const int s_Q14S_8 = 1024;
+extern const unsigned short t_Q14S_8[2032];
+const unsigned short t_Q14S_8[2032] = {
+  0x4000,0x0000 ,0x4000,0x0000 ,0x4000,0x0000 ,
+  0x22a3,0x187e ,0x3249,0x0c7c ,0x11a8,0x238e ,
+  0x0000,0x2d41 ,0x22a3,0x187e ,0xdd5d,0x3b21 ,
+  0xdd5d,0x3b21 ,0x11a8,0x238e ,0xb4be,0x3ec5 ,
+  0xc000,0x4000 ,0x0000,0x2d41 ,0xa57e,0x2d41 ,
+  0xac61,0x3b21 ,0xee58,0x3537 ,0xb4be,0x0c7c ,
+  0xa57e,0x2d41 ,0xdd5d,0x3b21 ,0xdd5d,0xe782 ,
+  0xac61,0x187e ,0xcdb7,0x3ec5 ,0x11a8,0xcac9 ,
+  0x4000,0x0000 ,0x4000,0x0000 ,0x4000,0x0000 ,
+  0x396b,0x0646 ,0x3cc8,0x0324 ,0x35eb,0x0964 ,
+  0x3249,0x0c7c ,0x396b,0x0646 ,0x2aaa,0x1294 ,
+  0x2aaa,0x1294 ,0x35eb,0x0964 ,0x1e7e,0x1b5d ,
+  0x22a3,0x187e ,0x3249,0x0c7c ,0x11a8,0x238e ,
+  0x1a46,0x1e2b ,0x2e88,0x0f8d ,0x0471,0x2afb ,
+  0x11a8,0x238e ,0x2aaa,0x1294 ,0xf721,0x3179 ,
+  0x08df,0x289a ,0x26b3,0x1590 ,0xea02,0x36e5 ,
+  0x0000,0x2d41 ,0x22a3,0x187e ,0xdd5d,0x3b21 ,
+  0xf721,0x3179 ,0x1e7e,0x1b5d ,0xd178,0x3e15 ,
+  0xee58,0x3537 ,0x1a46,0x1e2b ,0xc695,0x3fb1 ,
+  0xe5ba,0x3871 ,0x15fe,0x20e7 ,0xbcf0,0x3fec ,
+  0xdd5d,0x3b21 ,0x11a8,0x238e ,0xb4be,0x3ec5 ,
+  0xd556,0x3d3f ,0x0d48,0x2620 ,0xae2e,0x3c42 ,
+  0xcdb7,0x3ec5 ,0x08df,0x289a ,0xa963,0x3871 ,
+  0xc695,0x3fb1 ,0x0471,0x2afb ,0xa678,0x3368 ,
+  0xc000,0x4000 ,0x0000,0x2d41 ,0xa57e,0x2d41 ,
+  0xba09,0x3fb1 ,0xfb8f,0x2f6c ,0xa678,0x2620 ,
+  0xb4be,0x3ec5 ,0xf721,0x3179 ,0xa963,0x1e2b ,
+  0xb02d,0x3d3f ,0xf2b8,0x3368 ,0xae2e,0x1590 ,
+  0xac61,0x3b21 ,0xee58,0x3537 ,0xb4be,0x0c7c ,
+  0xa963,0x3871 ,0xea02,0x36e5 ,0xbcf0,0x0324 ,
+  0xa73b,0x3537 ,0xe5ba,0x3871 ,0xc695,0xf9ba ,
+  0xa5ed,0x3179 ,0xe182,0x39db ,0xd178,0xf073 ,
+  0xa57e,0x2d41 ,0xdd5d,0x3b21 ,0xdd5d,0xe782 ,
+  0xa5ed,0x289a ,0xd94d,0x3c42 ,0xea02,0xdf19 ,
+  0xa73b,0x238e ,0xd556,0x3d3f ,0xf721,0xd766 ,
+  0xa963,0x1e2b ,0xd178,0x3e15 ,0x0471,0xd094 ,
+  0xac61,0x187e ,0xcdb7,0x3ec5 ,0x11a8,0xcac9 ,
+  0xb02d,0x1294 ,0xca15,0x3f4f ,0x1e7e,0xc625 ,
+  0xb4be,0x0c7c ,0xc695,0x3fb1 ,0x2aaa,0xc2c1 ,
+  0xba09,0x0646 ,0xc338,0x3fec ,0x35eb,0xc0b1 ,
+  0x4000,0x0000 ,0x4000,0x0000 ,0x4000,0x0000 ,
+  0x3e69,0x0192 ,0x3f36,0x00c9 ,0x3d9a,0x025b ,
+  0x3cc8,0x0324 ,0x3e69,0x0192 ,0x3b1e,0x04b5 ,
+  0x3b1e,0x04b5 ,0x3d9a,0x025b ,0x388e,0x070e ,
+  0x396b,0x0646 ,0x3cc8,0x0324 ,0x35eb,0x0964 ,
+  0x37af,0x07d6 ,0x3bf4,0x03ed ,0x3334,0x0bb7 ,
+  0x35eb,0x0964 ,0x3b1e,0x04b5 ,0x306c,0x0e06 ,
+  0x341e,0x0af1 ,0x3a46,0x057e ,0x2d93,0x1050 ,
+  0x3249,0x0c7c ,0x396b,0x0646 ,0x2aaa,0x1294 ,
+  0x306c,0x0e06 ,0x388e,0x070e ,0x27b3,0x14d2 ,
+  0x2e88,0x0f8d ,0x37af,0x07d6 ,0x24ae,0x1709 ,
+  0x2c9d,0x1112 ,0x36ce,0x089d ,0x219c,0x1937 ,
+  0x2aaa,0x1294 ,0x35eb,0x0964 ,0x1e7e,0x1b5d ,
+  0x28b2,0x1413 ,0x3505,0x0a2b ,0x1b56,0x1d79 ,
+  0x26b3,0x1590 ,0x341e,0x0af1 ,0x1824,0x1f8c ,
+  0x24ae,0x1709 ,0x3334,0x0bb7 ,0x14ea,0x2193 ,
+  0x22a3,0x187e ,0x3249,0x0c7c ,0x11a8,0x238e ,
+  0x2093,0x19ef ,0x315b,0x0d41 ,0x0e61,0x257e ,
+  0x1e7e,0x1b5d ,0x306c,0x0e06 ,0x0b14,0x2760 ,
+  0x1c64,0x1cc6 ,0x2f7b,0x0eca ,0x07c4,0x2935 ,
+  0x1a46,0x1e2b ,0x2e88,0x0f8d ,0x0471,0x2afb ,
+  0x1824,0x1f8c ,0x2d93,0x1050 ,0x011c,0x2cb2 ,
+  0x15fe,0x20e7 ,0x2c9d,0x1112 ,0xfdc7,0x2e5a ,
+  0x13d5,0x223d ,0x2ba4,0x11d3 ,0xfa73,0x2ff2 ,
+  0x11a8,0x238e ,0x2aaa,0x1294 ,0xf721,0x3179 ,
+  0x0f79,0x24da ,0x29af,0x1354 ,0xf3d2,0x32ef ,
+  0x0d48,0x2620 ,0x28b2,0x1413 ,0xf087,0x3453 ,
+  0x0b14,0x2760 ,0x27b3,0x14d2 ,0xed41,0x35a5 ,
+  0x08df,0x289a ,0x26b3,0x1590 ,0xea02,0x36e5 ,
+  0x06a9,0x29ce ,0x25b1,0x164c ,0xe6cb,0x3812 ,
+  0x0471,0x2afb ,0x24ae,0x1709 ,0xe39c,0x392b ,
+  0x0239,0x2c21 ,0x23a9,0x17c4 ,0xe077,0x3a30 ,
+  0x0000,0x2d41 ,0x22a3,0x187e ,0xdd5d,0x3b21 ,
+  0xfdc7,0x2e5a ,0x219c,0x1937 ,0xda4f,0x3bfd ,
+  0xfb8f,0x2f6c ,0x2093,0x19ef ,0xd74e,0x3cc5 ,
+  0xf957,0x3076 ,0x1f89,0x1aa7 ,0xd45c,0x3d78 ,
+  0xf721,0x3179 ,0x1e7e,0x1b5d ,0xd178,0x3e15 ,
+  0xf4ec,0x3274 ,0x1d72,0x1c12 ,0xcea5,0x3e9d ,
+  0xf2b8,0x3368 ,0x1c64,0x1cc6 ,0xcbe2,0x3f0f ,
+  0xf087,0x3453 ,0x1b56,0x1d79 ,0xc932,0x3f6b ,
+  0xee58,0x3537 ,0x1a46,0x1e2b ,0xc695,0x3fb1 ,
+  0xec2b,0x3612 ,0x1935,0x1edc ,0xc40c,0x3fe1 ,
+  0xea02,0x36e5 ,0x1824,0x1f8c ,0xc197,0x3ffb ,
+  0xe7dc,0x37b0 ,0x1711,0x203a ,0xbf38,0x3fff ,
+  0xe5ba,0x3871 ,0x15fe,0x20e7 ,0xbcf0,0x3fec ,
+  0xe39c,0x392b ,0x14ea,0x2193 ,0xbabf,0x3fc4 ,
+  0xe182,0x39db ,0x13d5,0x223d ,0xb8a6,0x3f85 ,
+  0xdf6d,0x3a82 ,0x12bf,0x22e7 ,0xb6a5,0x3f30 ,
+  0xdd5d,0x3b21 ,0x11a8,0x238e ,0xb4be,0x3ec5 ,
+  0xdb52,0x3bb6 ,0x1091,0x2435 ,0xb2f2,0x3e45 ,
+  0xd94d,0x3c42 ,0x0f79,0x24da ,0xb140,0x3daf ,
+  0xd74e,0x3cc5 ,0x0e61,0x257e ,0xafa9,0x3d03 ,
+  0xd556,0x3d3f ,0x0d48,0x2620 ,0xae2e,0x3c42 ,
+  0xd363,0x3daf ,0x0c2e,0x26c1 ,0xacd0,0x3b6d ,
+  0xd178,0x3e15 ,0x0b14,0x2760 ,0xab8e,0x3a82 ,
+  0xcf94,0x3e72 ,0x09fa,0x27fe ,0xaa6a,0x3984 ,
+  0xcdb7,0x3ec5 ,0x08df,0x289a ,0xa963,0x3871 ,
+  0xcbe2,0x3f0f ,0x07c4,0x2935 ,0xa87b,0x374b ,
+  0xca15,0x3f4f ,0x06a9,0x29ce ,0xa7b1,0x3612 ,
+  0xc851,0x3f85 ,0x058d,0x2a65 ,0xa705,0x34c6 ,
+  0xc695,0x3fb1 ,0x0471,0x2afb ,0xa678,0x3368 ,
+  0xc4e2,0x3fd4 ,0x0355,0x2b8f ,0xa60b,0x31f8 ,
+  0xc338,0x3fec ,0x0239,0x2c21 ,0xa5bc,0x3076 ,
+  0xc197,0x3ffb ,0x011c,0x2cb2 ,0xa58d,0x2ee4 ,
+  0xc000,0x4000 ,0x0000,0x2d41 ,0xa57e,0x2d41 ,
+  0xbe73,0x3ffb ,0xfee4,0x2dcf ,0xa58d,0x2b8f ,
+  0xbcf0,0x3fec ,0xfdc7,0x2e5a ,0xa5bc,0x29ce ,
+  0xbb77,0x3fd4 ,0xfcab,0x2ee4 ,0xa60b,0x27fe ,
+  0xba09,0x3fb1 ,0xfb8f,0x2f6c ,0xa678,0x2620 ,
+  0xb8a6,0x3f85 ,0xfa73,0x2ff2 ,0xa705,0x2435 ,
+  0xb74d,0x3f4f ,0xf957,0x3076 ,0xa7b1,0x223d ,
+  0xb600,0x3f0f ,0xf83c,0x30f9 ,0xa87b,0x203a ,
+  0xb4be,0x3ec5 ,0xf721,0x3179 ,0xa963,0x1e2b ,
+  0xb388,0x3e72 ,0xf606,0x31f8 ,0xaa6a,0x1c12 ,
+  0xb25e,0x3e15 ,0xf4ec,0x3274 ,0xab8e,0x19ef ,
+  0xb140,0x3daf ,0xf3d2,0x32ef ,0xacd0,0x17c4 ,
+  0xb02d,0x3d3f ,0xf2b8,0x3368 ,0xae2e,0x1590 ,
+  0xaf28,0x3cc5 ,0xf19f,0x33df ,0xafa9,0x1354 ,
+  0xae2e,0x3c42 ,0xf087,0x3453 ,0xb140,0x1112 ,
+  0xad41,0x3bb6 ,0xef6f,0x34c6 ,0xb2f2,0x0eca ,
+  0xac61,0x3b21 ,0xee58,0x3537 ,0xb4be,0x0c7c ,
+  0xab8e,0x3a82 ,0xed41,0x35a5 ,0xb6a5,0x0a2b ,
+  0xaac8,0x39db ,0xec2b,0x3612 ,0xb8a6,0x07d6 ,
+  0xaa0f,0x392b ,0xeb16,0x367d ,0xbabf,0x057e ,
+  0xa963,0x3871 ,0xea02,0x36e5 ,0xbcf0,0x0324 ,
+  0xa8c5,0x37b0 ,0xe8ef,0x374b ,0xbf38,0x00c9 ,
+  0xa834,0x36e5 ,0xe7dc,0x37b0 ,0xc197,0xfe6e ,
+  0xa7b1,0x3612 ,0xe6cb,0x3812 ,0xc40c,0xfc13 ,
+  0xa73b,0x3537 ,0xe5ba,0x3871 ,0xc695,0xf9ba ,
+  0xa6d3,0x3453 ,0xe4aa,0x38cf ,0xc932,0xf763 ,
+  0xa678,0x3368 ,0xe39c,0x392b ,0xcbe2,0xf50f ,
+  0xa62c,0x3274 ,0xe28e,0x3984 ,0xcea5,0xf2bf ,
+  0xa5ed,0x3179 ,0xe182,0x39db ,0xd178,0xf073 ,
+  0xa5bc,0x3076 ,0xe077,0x3a30 ,0xd45c,0xee2d ,
+  0xa599,0x2f6c ,0xdf6d,0x3a82 ,0xd74e,0xebed ,
+  0xa585,0x2e5a ,0xde64,0x3ad3 ,0xda4f,0xe9b4 ,
+  0xa57e,0x2d41 ,0xdd5d,0x3b21 ,0xdd5d,0xe782 ,
+  0xa585,0x2c21 ,0xdc57,0x3b6d ,0xe077,0xe559 ,
+  0xa599,0x2afb ,0xdb52,0x3bb6 ,0xe39c,0xe33a ,
+  0xa5bc,0x29ce ,0xda4f,0x3bfd ,0xe6cb,0xe124 ,
+  0xa5ed,0x289a ,0xd94d,0x3c42 ,0xea02,0xdf19 ,
+  0xa62c,0x2760 ,0xd84d,0x3c85 ,0xed41,0xdd19 ,
+  0xa678,0x2620 ,0xd74e,0x3cc5 ,0xf087,0xdb26 ,
+  0xa6d3,0x24da ,0xd651,0x3d03 ,0xf3d2,0xd93f ,
+  0xa73b,0x238e ,0xd556,0x3d3f ,0xf721,0xd766 ,
+  0xa7b1,0x223d ,0xd45c,0x3d78 ,0xfa73,0xd59b ,
+  0xa834,0x20e7 ,0xd363,0x3daf ,0xfdc7,0xd3df ,
+  0xa8c5,0x1f8c ,0xd26d,0x3de3 ,0x011c,0xd231 ,
+  0xa963,0x1e2b ,0xd178,0x3e15 ,0x0471,0xd094 ,
+  0xaa0f,0x1cc6 ,0xd085,0x3e45 ,0x07c4,0xcf07 ,
+  0xaac8,0x1b5d ,0xcf94,0x3e72 ,0x0b14,0xcd8c ,
+  0xab8e,0x19ef ,0xcea5,0x3e9d ,0x0e61,0xcc21 ,
+  0xac61,0x187e ,0xcdb7,0x3ec5 ,0x11a8,0xcac9 ,
+  0xad41,0x1709 ,0xcccc,0x3eeb ,0x14ea,0xc983 ,
+  0xae2e,0x1590 ,0xcbe2,0x3f0f ,0x1824,0xc850 ,
+  0xaf28,0x1413 ,0xcafb,0x3f30 ,0x1b56,0xc731 ,
+  0xb02d,0x1294 ,0xca15,0x3f4f ,0x1e7e,0xc625 ,
+  0xb140,0x1112 ,0xc932,0x3f6b ,0x219c,0xc52d ,
+  0xb25e,0x0f8d ,0xc851,0x3f85 ,0x24ae,0xc44a ,
+  0xb388,0x0e06 ,0xc772,0x3f9c ,0x27b3,0xc37b ,
+  0xb4be,0x0c7c ,0xc695,0x3fb1 ,0x2aaa,0xc2c1 ,
+  0xb600,0x0af1 ,0xc5ba,0x3fc4 ,0x2d93,0xc21d ,
+  0xb74d,0x0964 ,0xc4e2,0x3fd4 ,0x306c,0xc18e ,
+  0xb8a6,0x07d6 ,0xc40c,0x3fe1 ,0x3334,0xc115 ,
+  0xba09,0x0646 ,0xc338,0x3fec ,0x35eb,0xc0b1 ,
+  0xbb77,0x04b5 ,0xc266,0x3ff5 ,0x388e,0xc064 ,
+  0xbcf0,0x0324 ,0xc197,0x3ffb ,0x3b1e,0xc02c ,
+  0xbe73,0x0192 ,0xc0ca,0x3fff ,0x3d9a,0xc00b ,
+  0x4000,0x0000 ,0x3f9b,0x0065 ,0x3f36,0x00c9 ,
+  0x3ed0,0x012e ,0x3e69,0x0192 ,0x3e02,0x01f7 ,
+  0x3d9a,0x025b ,0x3d31,0x02c0 ,0x3cc8,0x0324 ,
+  0x3c5f,0x0388 ,0x3bf4,0x03ed ,0x3b8a,0x0451 ,
+  0x3b1e,0x04b5 ,0x3ab2,0x051a ,0x3a46,0x057e ,
+  0x39d9,0x05e2 ,0x396b,0x0646 ,0x38fd,0x06aa ,
+  0x388e,0x070e ,0x381f,0x0772 ,0x37af,0x07d6 ,
+  0x373f,0x0839 ,0x36ce,0x089d ,0x365d,0x0901 ,
+  0x35eb,0x0964 ,0x3578,0x09c7 ,0x3505,0x0a2b ,
+  0x3492,0x0a8e ,0x341e,0x0af1 ,0x33a9,0x0b54 ,
+  0x3334,0x0bb7 ,0x32bf,0x0c1a ,0x3249,0x0c7c ,
+  0x31d2,0x0cdf ,0x315b,0x0d41 ,0x30e4,0x0da4 ,
+  0x306c,0x0e06 ,0x2ff4,0x0e68 ,0x2f7b,0x0eca ,
+  0x2f02,0x0f2b ,0x2e88,0x0f8d ,0x2e0e,0x0fee ,
+  0x2d93,0x1050 ,0x2d18,0x10b1 ,0x2c9d,0x1112 ,
+  0x2c21,0x1173 ,0x2ba4,0x11d3 ,0x2b28,0x1234 ,
+  0x2aaa,0x1294 ,0x2a2d,0x12f4 ,0x29af,0x1354 ,
+  0x2931,0x13b4 ,0x28b2,0x1413 ,0x2833,0x1473 ,
+  0x27b3,0x14d2 ,0x2733,0x1531 ,0x26b3,0x1590 ,
+  0x2632,0x15ee ,0x25b1,0x164c ,0x252f,0x16ab ,
+  0x24ae,0x1709 ,0x242b,0x1766 ,0x23a9,0x17c4 ,
+  0x2326,0x1821 ,0x22a3,0x187e ,0x221f,0x18db ,
+  0x219c,0x1937 ,0x2117,0x1993 ,0x2093,0x19ef ,
+  0x200e,0x1a4b ,0x1f89,0x1aa7 ,0x1f04,0x1b02 ,
+  0x1e7e,0x1b5d ,0x1df8,0x1bb8 ,0x1d72,0x1c12 ,
+  0x1ceb,0x1c6c ,0x1c64,0x1cc6 ,0x1bdd,0x1d20 ,
+  0x1b56,0x1d79 ,0x1ace,0x1dd3 ,0x1a46,0x1e2b ,
+  0x19be,0x1e84 ,0x1935,0x1edc ,0x18ad,0x1f34 ,
+  0x1824,0x1f8c ,0x179b,0x1fe3 ,0x1711,0x203a ,
+  0x1688,0x2091 ,0x15fe,0x20e7 ,0x1574,0x213d ,
+  0x14ea,0x2193 ,0x145f,0x21e8 ,0x13d5,0x223d ,
+  0x134a,0x2292 ,0x12bf,0x22e7 ,0x1234,0x233b ,
+  0x11a8,0x238e ,0x111d,0x23e2 ,0x1091,0x2435 ,
+  0x1005,0x2488 ,0x0f79,0x24da ,0x0eed,0x252c ,
+  0x0e61,0x257e ,0x0dd4,0x25cf ,0x0d48,0x2620 ,
+  0x0cbb,0x2671 ,0x0c2e,0x26c1 ,0x0ba1,0x2711 ,
+  0x0b14,0x2760 ,0x0a87,0x27af ,0x09fa,0x27fe ,
+  0x096d,0x284c ,0x08df,0x289a ,0x0852,0x28e7 ,
+  0x07c4,0x2935 ,0x0736,0x2981 ,0x06a9,0x29ce ,
+  0x061b,0x2a1a ,0x058d,0x2a65 ,0x04ff,0x2ab0 ,
+  0x0471,0x2afb ,0x03e3,0x2b45 ,0x0355,0x2b8f ,
+  0x02c7,0x2bd8 ,0x0239,0x2c21 ,0x01aa,0x2c6a ,
+  0x011c,0x2cb2 ,0x008e,0x2cfa ,0x0000,0x2d41 ,
+  0xff72,0x2d88 ,0xfee4,0x2dcf ,0xfe56,0x2e15 ,
+  0xfdc7,0x2e5a ,0xfd39,0x2e9f ,0xfcab,0x2ee4 ,
+  0xfc1d,0x2f28 ,0xfb8f,0x2f6c ,0xfb01,0x2faf ,
+  0xfa73,0x2ff2 ,0xf9e5,0x3034 ,0xf957,0x3076 ,
+  0xf8ca,0x30b8 ,0xf83c,0x30f9 ,0xf7ae,0x3139 ,
+  0xf721,0x3179 ,0xf693,0x31b9 ,0xf606,0x31f8 ,
+  0xf579,0x3236 ,0xf4ec,0x3274 ,0xf45f,0x32b2 ,
+  0xf3d2,0x32ef ,0xf345,0x332c ,0xf2b8,0x3368 ,
+  0xf22c,0x33a3 ,0xf19f,0x33df ,0xf113,0x3419 ,
+  0xf087,0x3453 ,0xeffb,0x348d ,0xef6f,0x34c6 ,
+  0xeee3,0x34ff ,0xee58,0x3537 ,0xedcc,0x356e ,
+  0xed41,0x35a5 ,0xecb6,0x35dc ,0xec2b,0x3612 ,
+  0xeba1,0x3648 ,0xeb16,0x367d ,0xea8c,0x36b1 ,
+  0xea02,0x36e5 ,0xe978,0x3718 ,0xe8ef,0x374b ,
+  0xe865,0x377e ,0xe7dc,0x37b0 ,0xe753,0x37e1 ,
+  0xe6cb,0x3812 ,0xe642,0x3842 ,0xe5ba,0x3871 ,
+  0xe532,0x38a1 ,0xe4aa,0x38cf ,0xe423,0x38fd ,
+  0xe39c,0x392b ,0xe315,0x3958 ,0xe28e,0x3984 ,
+  0xe208,0x39b0 ,0xe182,0x39db ,0xe0fc,0x3a06 ,
+  0xe077,0x3a30 ,0xdff2,0x3a59 ,0xdf6d,0x3a82 ,
+  0xdee9,0x3aab ,0xde64,0x3ad3 ,0xdde1,0x3afa ,
+  0xdd5d,0x3b21 ,0xdcda,0x3b47 ,0xdc57,0x3b6d ,
+  0xdbd5,0x3b92 ,0xdb52,0x3bb6 ,0xdad1,0x3bda ,
+  0xda4f,0x3bfd ,0xd9ce,0x3c20 ,0xd94d,0x3c42 ,
+  0xd8cd,0x3c64 ,0xd84d,0x3c85 ,0xd7cd,0x3ca5 ,
+  0xd74e,0x3cc5 ,0xd6cf,0x3ce4 ,0xd651,0x3d03 ,
+  0xd5d3,0x3d21 ,0xd556,0x3d3f ,0xd4d8,0x3d5b ,
+  0xd45c,0x3d78 ,0xd3df,0x3d93 ,0xd363,0x3daf ,
+  0xd2e8,0x3dc9 ,0xd26d,0x3de3 ,0xd1f2,0x3dfc ,
+  0xd178,0x3e15 ,0xd0fe,0x3e2d ,0xd085,0x3e45 ,
+  0xd00c,0x3e5c ,0xcf94,0x3e72 ,0xcf1c,0x3e88 ,
+  0xcea5,0x3e9d ,0xce2e,0x3eb1 ,0xcdb7,0x3ec5 ,
+  0xcd41,0x3ed8 ,0xcccc,0x3eeb ,0xcc57,0x3efd ,
+  0xcbe2,0x3f0f ,0xcb6e,0x3f20 ,0xcafb,0x3f30 ,
+  0xca88,0x3f40 ,0xca15,0x3f4f ,0xc9a3,0x3f5d ,
+  0xc932,0x3f6b ,0xc8c1,0x3f78 ,0xc851,0x3f85 ,
+  0xc7e1,0x3f91 ,0xc772,0x3f9c ,0xc703,0x3fa7 ,
+  0xc695,0x3fb1 ,0xc627,0x3fbb ,0xc5ba,0x3fc4 ,
+  0xc54e,0x3fcc ,0xc4e2,0x3fd4 ,0xc476,0x3fdb ,
+  0xc40c,0x3fe1 ,0xc3a1,0x3fe7 ,0xc338,0x3fec ,
+  0xc2cf,0x3ff1 ,0xc266,0x3ff5 ,0xc1fe,0x3ff8 ,
+  0xc197,0x3ffb ,0xc130,0x3ffd ,0xc0ca,0x3fff ,
+  0xc065,0x4000 ,0xc000,0x4000 ,0xbf9c,0x4000 ,
+  0xbf38,0x3fff ,0xbed5,0x3ffd ,0xbe73,0x3ffb ,
+  0xbe11,0x3ff8 ,0xbdb0,0x3ff5 ,0xbd50,0x3ff1 ,
+  0xbcf0,0x3fec ,0xbc91,0x3fe7 ,0xbc32,0x3fe1 ,
+  0xbbd4,0x3fdb ,0xbb77,0x3fd4 ,0xbb1b,0x3fcc ,
+  0xbabf,0x3fc4 ,0xba64,0x3fbb ,0xba09,0x3fb1 ,
+  0xb9af,0x3fa7 ,0xb956,0x3f9c ,0xb8fd,0x3f91 ,
+  0xb8a6,0x3f85 ,0xb84f,0x3f78 ,0xb7f8,0x3f6b ,
+  0xb7a2,0x3f5d ,0xb74d,0x3f4f ,0xb6f9,0x3f40 ,
+  0xb6a5,0x3f30 ,0xb652,0x3f20 ,0xb600,0x3f0f ,
+  0xb5af,0x3efd ,0xb55e,0x3eeb ,0xb50e,0x3ed8 ,
+  0xb4be,0x3ec5 ,0xb470,0x3eb1 ,0xb422,0x3e9d ,
+  0xb3d5,0x3e88 ,0xb388,0x3e72 ,0xb33d,0x3e5c ,
+  0xb2f2,0x3e45 ,0xb2a7,0x3e2d ,0xb25e,0x3e15 ,
+  0xb215,0x3dfc ,0xb1cd,0x3de3 ,0xb186,0x3dc9 ,
+  0xb140,0x3daf ,0xb0fa,0x3d93 ,0xb0b5,0x3d78 ,
+  0xb071,0x3d5b ,0xb02d,0x3d3f ,0xafeb,0x3d21 ,
+  0xafa9,0x3d03 ,0xaf68,0x3ce4 ,0xaf28,0x3cc5 ,
+  0xaee8,0x3ca5 ,0xaea9,0x3c85 ,0xae6b,0x3c64 ,
+  0xae2e,0x3c42 ,0xadf2,0x3c20 ,0xadb6,0x3bfd ,
+  0xad7b,0x3bda ,0xad41,0x3bb6 ,0xad08,0x3b92 ,
+  0xacd0,0x3b6d ,0xac98,0x3b47 ,0xac61,0x3b21 ,
+  0xac2b,0x3afa ,0xabf6,0x3ad3 ,0xabc2,0x3aab ,
+  0xab8e,0x3a82 ,0xab5b,0x3a59 ,0xab29,0x3a30 ,
+  0xaaf8,0x3a06 ,0xaac8,0x39db ,0xaa98,0x39b0 ,
+  0xaa6a,0x3984 ,0xaa3c,0x3958 ,0xaa0f,0x392b ,
+  0xa9e3,0x38fd ,0xa9b7,0x38cf ,0xa98d,0x38a1 ,
+  0xa963,0x3871 ,0xa93a,0x3842 ,0xa912,0x3812 ,
+  0xa8eb,0x37e1 ,0xa8c5,0x37b0 ,0xa89f,0x377e ,
+  0xa87b,0x374b ,0xa857,0x3718 ,0xa834,0x36e5 ,
+  0xa812,0x36b1 ,0xa7f1,0x367d ,0xa7d0,0x3648 ,
+  0xa7b1,0x3612 ,0xa792,0x35dc ,0xa774,0x35a5 ,
+  0xa757,0x356e ,0xa73b,0x3537 ,0xa71f,0x34ff ,
+  0xa705,0x34c6 ,0xa6eb,0x348d ,0xa6d3,0x3453 ,
+  0xa6bb,0x3419 ,0xa6a4,0x33df ,0xa68e,0x33a3 ,
+  0xa678,0x3368 ,0xa664,0x332c ,0xa650,0x32ef ,
+  0xa63e,0x32b2 ,0xa62c,0x3274 ,0xa61b,0x3236 ,
+  0xa60b,0x31f8 ,0xa5fb,0x31b9 ,0xa5ed,0x3179 ,
+  0xa5e0,0x3139 ,0xa5d3,0x30f9 ,0xa5c7,0x30b8 ,
+  0xa5bc,0x3076 ,0xa5b2,0x3034 ,0xa5a9,0x2ff2 ,
+  0xa5a1,0x2faf ,0xa599,0x2f6c ,0xa593,0x2f28 ,
+  0xa58d,0x2ee4 ,0xa588,0x2e9f ,0xa585,0x2e5a ,
+  0xa581,0x2e15 ,0xa57f,0x2dcf ,0xa57e,0x2d88 ,
+  0xa57e,0x2d41 ,0xa57e,0x2cfa ,0xa57f,0x2cb2 ,
+  0xa581,0x2c6a ,0xa585,0x2c21 ,0xa588,0x2bd8 ,
+  0xa58d,0x2b8f ,0xa593,0x2b45 ,0xa599,0x2afb ,
+  0xa5a1,0x2ab0 ,0xa5a9,0x2a65 ,0xa5b2,0x2a1a ,
+  0xa5bc,0x29ce ,0xa5c7,0x2981 ,0xa5d3,0x2935 ,
+  0xa5e0,0x28e7 ,0xa5ed,0x289a ,0xa5fb,0x284c ,
+  0xa60b,0x27fe ,0xa61b,0x27af ,0xa62c,0x2760 ,
+  0xa63e,0x2711 ,0xa650,0x26c1 ,0xa664,0x2671 ,
+  0xa678,0x2620 ,0xa68e,0x25cf ,0xa6a4,0x257e ,
+  0xa6bb,0x252c ,0xa6d3,0x24da ,0xa6eb,0x2488 ,
+  0xa705,0x2435 ,0xa71f,0x23e2 ,0xa73b,0x238e ,
+  0xa757,0x233b ,0xa774,0x22e7 ,0xa792,0x2292 ,
+  0xa7b1,0x223d ,0xa7d0,0x21e8 ,0xa7f1,0x2193 ,
+  0xa812,0x213d ,0xa834,0x20e7 ,0xa857,0x2091 ,
+  0xa87b,0x203a ,0xa89f,0x1fe3 ,0xa8c5,0x1f8c ,
+  0xa8eb,0x1f34 ,0xa912,0x1edc ,0xa93a,0x1e84 ,
+  0xa963,0x1e2b ,0xa98d,0x1dd3 ,0xa9b7,0x1d79 ,
+  0xa9e3,0x1d20 ,0xaa0f,0x1cc6 ,0xaa3c,0x1c6c ,
+  0xaa6a,0x1c12 ,0xaa98,0x1bb8 ,0xaac8,0x1b5d ,
+  0xaaf8,0x1b02 ,0xab29,0x1aa7 ,0xab5b,0x1a4b ,
+  0xab8e,0x19ef ,0xabc2,0x1993 ,0xabf6,0x1937 ,
+  0xac2b,0x18db ,0xac61,0x187e ,0xac98,0x1821 ,
+  0xacd0,0x17c4 ,0xad08,0x1766 ,0xad41,0x1709 ,
+  0xad7b,0x16ab ,0xadb6,0x164c ,0xadf2,0x15ee ,
+  0xae2e,0x1590 ,0xae6b,0x1531 ,0xaea9,0x14d2 ,
+  0xaee8,0x1473 ,0xaf28,0x1413 ,0xaf68,0x13b4 ,
+  0xafa9,0x1354 ,0xafeb,0x12f4 ,0xb02d,0x1294 ,
+  0xb071,0x1234 ,0xb0b5,0x11d3 ,0xb0fa,0x1173 ,
+  0xb140,0x1112 ,0xb186,0x10b1 ,0xb1cd,0x1050 ,
+  0xb215,0x0fee ,0xb25e,0x0f8d ,0xb2a7,0x0f2b ,
+  0xb2f2,0x0eca ,0xb33d,0x0e68 ,0xb388,0x0e06 ,
+  0xb3d5,0x0da4 ,0xb422,0x0d41 ,0xb470,0x0cdf ,
+  0xb4be,0x0c7c ,0xb50e,0x0c1a ,0xb55e,0x0bb7 ,
+  0xb5af,0x0b54 ,0xb600,0x0af1 ,0xb652,0x0a8e ,
+  0xb6a5,0x0a2b ,0xb6f9,0x09c7 ,0xb74d,0x0964 ,
+  0xb7a2,0x0901 ,0xb7f8,0x089d ,0xb84f,0x0839 ,
+  0xb8a6,0x07d6 ,0xb8fd,0x0772 ,0xb956,0x070e ,
+  0xb9af,0x06aa ,0xba09,0x0646 ,0xba64,0x05e2 ,
+  0xbabf,0x057e ,0xbb1b,0x051a ,0xbb77,0x04b5 ,
+  0xbbd4,0x0451 ,0xbc32,0x03ed ,0xbc91,0x0388 ,
+  0xbcf0,0x0324 ,0xbd50,0x02c0 ,0xbdb0,0x025b ,
+  0xbe11,0x01f7 ,0xbe73,0x0192 ,0xbed5,0x012e ,
+  0xbf38,0x00c9 ,0xbf9c,0x0065 };
+
+
+extern const int s_Q14R_8;
+const int s_Q14R_8 = 1024;
+extern const unsigned short t_Q14R_8[2032];
+const unsigned short t_Q14R_8[2032] = {
+  0x4000,0x0000 ,0x4000,0x0000 ,0x4000,0x0000 ,
+  0x3b21,0x187e ,0x3ec5,0x0c7c ,0x3537,0x238e ,
+  0x2d41,0x2d41 ,0x3b21,0x187e ,0x187e,0x3b21 ,
+  0x187e,0x3b21 ,0x3537,0x238e ,0xf384,0x3ec5 ,
+  0x0000,0x4000 ,0x2d41,0x2d41 ,0xd2bf,0x2d41 ,
+  0xe782,0x3b21 ,0x238e,0x3537 ,0xc13b,0x0c7c ,
+  0xd2bf,0x2d41 ,0x187e,0x3b21 ,0xc4df,0xe782 ,
+  0xc4df,0x187e ,0x0c7c,0x3ec5 ,0xdc72,0xcac9 ,
+  0x4000,0x0000 ,0x4000,0x0000 ,0x4000,0x0000 ,
+  0x3fb1,0x0646 ,0x3fec,0x0324 ,0x3f4f,0x0964 ,
+  0x3ec5,0x0c7c ,0x3fb1,0x0646 ,0x3d3f,0x1294 ,
+  0x3d3f,0x1294 ,0x3f4f,0x0964 ,0x39db,0x1b5d ,
+  0x3b21,0x187e ,0x3ec5,0x0c7c ,0x3537,0x238e ,
+  0x3871,0x1e2b ,0x3e15,0x0f8d ,0x2f6c,0x2afb ,
+  0x3537,0x238e ,0x3d3f,0x1294 ,0x289a,0x3179 ,
+  0x3179,0x289a ,0x3c42,0x1590 ,0x20e7,0x36e5 ,
+  0x2d41,0x2d41 ,0x3b21,0x187e ,0x187e,0x3b21 ,
+  0x289a,0x3179 ,0x39db,0x1b5d ,0x0f8d,0x3e15 ,
+  0x238e,0x3537 ,0x3871,0x1e2b ,0x0646,0x3fb1 ,
+  0x1e2b,0x3871 ,0x36e5,0x20e7 ,0xfcdc,0x3fec ,
+  0x187e,0x3b21 ,0x3537,0x238e ,0xf384,0x3ec5 ,
+  0x1294,0x3d3f ,0x3368,0x2620 ,0xea70,0x3c42 ,
+  0x0c7c,0x3ec5 ,0x3179,0x289a ,0xe1d5,0x3871 ,
+  0x0646,0x3fb1 ,0x2f6c,0x2afb ,0xd9e0,0x3368 ,
+  0x0000,0x4000 ,0x2d41,0x2d41 ,0xd2bf,0x2d41 ,
+  0xf9ba,0x3fb1 ,0x2afb,0x2f6c ,0xcc98,0x2620 ,
+  0xf384,0x3ec5 ,0x289a,0x3179 ,0xc78f,0x1e2b ,
+  0xed6c,0x3d3f ,0x2620,0x3368 ,0xc3be,0x1590 ,
+  0xe782,0x3b21 ,0x238e,0x3537 ,0xc13b,0x0c7c ,
+  0xe1d5,0x3871 ,0x20e7,0x36e5 ,0xc014,0x0324 ,
+  0xdc72,0x3537 ,0x1e2b,0x3871 ,0xc04f,0xf9ba ,
+  0xd766,0x3179 ,0x1b5d,0x39db ,0xc1eb,0xf073 ,
+  0xd2bf,0x2d41 ,0x187e,0x3b21 ,0xc4df,0xe782 ,
+  0xce87,0x289a ,0x1590,0x3c42 ,0xc91b,0xdf19 ,
+  0xcac9,0x238e ,0x1294,0x3d3f ,0xce87,0xd766 ,
+  0xc78f,0x1e2b ,0x0f8d,0x3e15 ,0xd505,0xd094 ,
+  0xc4df,0x187e ,0x0c7c,0x3ec5 ,0xdc72,0xcac9 ,
+  0xc2c1,0x1294 ,0x0964,0x3f4f ,0xe4a3,0xc625 ,
+  0xc13b,0x0c7c ,0x0646,0x3fb1 ,0xed6c,0xc2c1 ,
+  0xc04f,0x0646 ,0x0324,0x3fec ,0xf69c,0xc0b1 ,
+  0x4000,0x0000 ,0x4000,0x0000 ,0x4000,0x0000 ,
+  0x3ffb,0x0192 ,0x3fff,0x00c9 ,0x3ff5,0x025b ,
+  0x3fec,0x0324 ,0x3ffb,0x0192 ,0x3fd4,0x04b5 ,
+  0x3fd4,0x04b5 ,0x3ff5,0x025b ,0x3f9c,0x070e ,
+  0x3fb1,0x0646 ,0x3fec,0x0324 ,0x3f4f,0x0964 ,
+  0x3f85,0x07d6 ,0x3fe1,0x03ed ,0x3eeb,0x0bb7 ,
+  0x3f4f,0x0964 ,0x3fd4,0x04b5 ,0x3e72,0x0e06 ,
+  0x3f0f,0x0af1 ,0x3fc4,0x057e ,0x3de3,0x1050 ,
+  0x3ec5,0x0c7c ,0x3fb1,0x0646 ,0x3d3f,0x1294 ,
+  0x3e72,0x0e06 ,0x3f9c,0x070e ,0x3c85,0x14d2 ,
+  0x3e15,0x0f8d ,0x3f85,0x07d6 ,0x3bb6,0x1709 ,
+  0x3daf,0x1112 ,0x3f6b,0x089d ,0x3ad3,0x1937 ,
+  0x3d3f,0x1294 ,0x3f4f,0x0964 ,0x39db,0x1b5d ,
+  0x3cc5,0x1413 ,0x3f30,0x0a2b ,0x38cf,0x1d79 ,
+  0x3c42,0x1590 ,0x3f0f,0x0af1 ,0x37b0,0x1f8c ,
+  0x3bb6,0x1709 ,0x3eeb,0x0bb7 ,0x367d,0x2193 ,
+  0x3b21,0x187e ,0x3ec5,0x0c7c ,0x3537,0x238e ,
+  0x3a82,0x19ef ,0x3e9d,0x0d41 ,0x33df,0x257e ,
+  0x39db,0x1b5d ,0x3e72,0x0e06 ,0x3274,0x2760 ,
+  0x392b,0x1cc6 ,0x3e45,0x0eca ,0x30f9,0x2935 ,
+  0x3871,0x1e2b ,0x3e15,0x0f8d ,0x2f6c,0x2afb ,
+  0x37b0,0x1f8c ,0x3de3,0x1050 ,0x2dcf,0x2cb2 ,
+  0x36e5,0x20e7 ,0x3daf,0x1112 ,0x2c21,0x2e5a ,
+  0x3612,0x223d ,0x3d78,0x11d3 ,0x2a65,0x2ff2 ,
+  0x3537,0x238e ,0x3d3f,0x1294 ,0x289a,0x3179 ,
+  0x3453,0x24da ,0x3d03,0x1354 ,0x26c1,0x32ef ,
+  0x3368,0x2620 ,0x3cc5,0x1413 ,0x24da,0x3453 ,
+  0x3274,0x2760 ,0x3c85,0x14d2 ,0x22e7,0x35a5 ,
+  0x3179,0x289a ,0x3c42,0x1590 ,0x20e7,0x36e5 ,
+  0x3076,0x29ce ,0x3bfd,0x164c ,0x1edc,0x3812 ,
+  0x2f6c,0x2afb ,0x3bb6,0x1709 ,0x1cc6,0x392b ,
+  0x2e5a,0x2c21 ,0x3b6d,0x17c4 ,0x1aa7,0x3a30 ,
+  0x2d41,0x2d41 ,0x3b21,0x187e ,0x187e,0x3b21 ,
+  0x2c21,0x2e5a ,0x3ad3,0x1937 ,0x164c,0x3bfd ,
+  0x2afb,0x2f6c ,0x3a82,0x19ef ,0x1413,0x3cc5 ,
+  0x29ce,0x3076 ,0x3a30,0x1aa7 ,0x11d3,0x3d78 ,
+  0x289a,0x3179 ,0x39db,0x1b5d ,0x0f8d,0x3e15 ,
+  0x2760,0x3274 ,0x3984,0x1c12 ,0x0d41,0x3e9d ,
+  0x2620,0x3368 ,0x392b,0x1cc6 ,0x0af1,0x3f0f ,
+  0x24da,0x3453 ,0x38cf,0x1d79 ,0x089d,0x3f6b ,
+  0x238e,0x3537 ,0x3871,0x1e2b ,0x0646,0x3fb1 ,
+  0x223d,0x3612 ,0x3812,0x1edc ,0x03ed,0x3fe1 ,
+  0x20e7,0x36e5 ,0x37b0,0x1f8c ,0x0192,0x3ffb ,
+  0x1f8c,0x37b0 ,0x374b,0x203a ,0xff37,0x3fff ,
+  0x1e2b,0x3871 ,0x36e5,0x20e7 ,0xfcdc,0x3fec ,
+  0x1cc6,0x392b ,0x367d,0x2193 ,0xfa82,0x3fc4 ,
+  0x1b5d,0x39db ,0x3612,0x223d ,0xf82a,0x3f85 ,
+  0x19ef,0x3a82 ,0x35a5,0x22e7 ,0xf5d5,0x3f30 ,
+  0x187e,0x3b21 ,0x3537,0x238e ,0xf384,0x3ec5 ,
+  0x1709,0x3bb6 ,0x34c6,0x2435 ,0xf136,0x3e45 ,
+  0x1590,0x3c42 ,0x3453,0x24da ,0xeeee,0x3daf ,
+  0x1413,0x3cc5 ,0x33df,0x257e ,0xecac,0x3d03 ,
+  0x1294,0x3d3f ,0x3368,0x2620 ,0xea70,0x3c42 ,
+  0x1112,0x3daf ,0x32ef,0x26c1 ,0xe83c,0x3b6d ,
+  0x0f8d,0x3e15 ,0x3274,0x2760 ,0xe611,0x3a82 ,
+  0x0e06,0x3e72 ,0x31f8,0x27fe ,0xe3ee,0x3984 ,
+  0x0c7c,0x3ec5 ,0x3179,0x289a ,0xe1d5,0x3871 ,
+  0x0af1,0x3f0f ,0x30f9,0x2935 ,0xdfc6,0x374b ,
+  0x0964,0x3f4f ,0x3076,0x29ce ,0xddc3,0x3612 ,
+  0x07d6,0x3f85 ,0x2ff2,0x2a65 ,0xdbcb,0x34c6 ,
+  0x0646,0x3fb1 ,0x2f6c,0x2afb ,0xd9e0,0x3368 ,
+  0x04b5,0x3fd4 ,0x2ee4,0x2b8f ,0xd802,0x31f8 ,
+  0x0324,0x3fec ,0x2e5a,0x2c21 ,0xd632,0x3076 ,
+  0x0192,0x3ffb ,0x2dcf,0x2cb2 ,0xd471,0x2ee4 ,
+  0x0000,0x4000 ,0x2d41,0x2d41 ,0xd2bf,0x2d41 ,
+  0xfe6e,0x3ffb ,0x2cb2,0x2dcf ,0xd11c,0x2b8f ,
+  0xfcdc,0x3fec ,0x2c21,0x2e5a ,0xcf8a,0x29ce ,
+  0xfb4b,0x3fd4 ,0x2b8f,0x2ee4 ,0xce08,0x27fe ,
+  0xf9ba,0x3fb1 ,0x2afb,0x2f6c ,0xcc98,0x2620 ,
+  0xf82a,0x3f85 ,0x2a65,0x2ff2 ,0xcb3a,0x2435 ,
+  0xf69c,0x3f4f ,0x29ce,0x3076 ,0xc9ee,0x223d ,
+  0xf50f,0x3f0f ,0x2935,0x30f9 ,0xc8b5,0x203a ,
+  0xf384,0x3ec5 ,0x289a,0x3179 ,0xc78f,0x1e2b ,
+  0xf1fa,0x3e72 ,0x27fe,0x31f8 ,0xc67c,0x1c12 ,
+  0xf073,0x3e15 ,0x2760,0x3274 ,0xc57e,0x19ef ,
+  0xeeee,0x3daf ,0x26c1,0x32ef ,0xc493,0x17c4 ,
+  0xed6c,0x3d3f ,0x2620,0x3368 ,0xc3be,0x1590 ,
+  0xebed,0x3cc5 ,0x257e,0x33df ,0xc2fd,0x1354 ,
+  0xea70,0x3c42 ,0x24da,0x3453 ,0xc251,0x1112 ,
+  0xe8f7,0x3bb6 ,0x2435,0x34c6 ,0xc1bb,0x0eca ,
+  0xe782,0x3b21 ,0x238e,0x3537 ,0xc13b,0x0c7c ,
+  0xe611,0x3a82 ,0x22e7,0x35a5 ,0xc0d0,0x0a2b ,
+  0xe4a3,0x39db ,0x223d,0x3612 ,0xc07b,0x07d6 ,
+  0xe33a,0x392b ,0x2193,0x367d ,0xc03c,0x057e ,
+  0xe1d5,0x3871 ,0x20e7,0x36e5 ,0xc014,0x0324 ,
+  0xe074,0x37b0 ,0x203a,0x374b ,0xc001,0x00c9 ,
+  0xdf19,0x36e5 ,0x1f8c,0x37b0 ,0xc005,0xfe6e ,
+  0xddc3,0x3612 ,0x1edc,0x3812 ,0xc01f,0xfc13 ,
+  0xdc72,0x3537 ,0x1e2b,0x3871 ,0xc04f,0xf9ba ,
+  0xdb26,0x3453 ,0x1d79,0x38cf ,0xc095,0xf763 ,
+  0xd9e0,0x3368 ,0x1cc6,0x392b ,0xc0f1,0xf50f ,
+  0xd8a0,0x3274 ,0x1c12,0x3984 ,0xc163,0xf2bf ,
+  0xd766,0x3179 ,0x1b5d,0x39db ,0xc1eb,0xf073 ,
+  0xd632,0x3076 ,0x1aa7,0x3a30 ,0xc288,0xee2d ,
+  0xd505,0x2f6c ,0x19ef,0x3a82 ,0xc33b,0xebed ,
+  0xd3df,0x2e5a ,0x1937,0x3ad3 ,0xc403,0xe9b4 ,
+  0xd2bf,0x2d41 ,0x187e,0x3b21 ,0xc4df,0xe782 ,
+  0xd1a6,0x2c21 ,0x17c4,0x3b6d ,0xc5d0,0xe559 ,
+  0xd094,0x2afb ,0x1709,0x3bb6 ,0xc6d5,0xe33a ,
+  0xcf8a,0x29ce ,0x164c,0x3bfd ,0xc7ee,0xe124 ,
+  0xce87,0x289a ,0x1590,0x3c42 ,0xc91b,0xdf19 ,
+  0xcd8c,0x2760 ,0x14d2,0x3c85 ,0xca5b,0xdd19 ,
+  0xcc98,0x2620 ,0x1413,0x3cc5 ,0xcbad,0xdb26 ,
+  0xcbad,0x24da ,0x1354,0x3d03 ,0xcd11,0xd93f ,
+  0xcac9,0x238e ,0x1294,0x3d3f ,0xce87,0xd766 ,
+  0xc9ee,0x223d ,0x11d3,0x3d78 ,0xd00e,0xd59b ,
+  0xc91b,0x20e7 ,0x1112,0x3daf ,0xd1a6,0xd3df ,
+  0xc850,0x1f8c ,0x1050,0x3de3 ,0xd34e,0xd231 ,
+  0xc78f,0x1e2b ,0x0f8d,0x3e15 ,0xd505,0xd094 ,
+  0xc6d5,0x1cc6 ,0x0eca,0x3e45 ,0xd6cb,0xcf07 ,
+  0xc625,0x1b5d ,0x0e06,0x3e72 ,0xd8a0,0xcd8c ,
+  0xc57e,0x19ef ,0x0d41,0x3e9d ,0xda82,0xcc21 ,
+  0xc4df,0x187e ,0x0c7c,0x3ec5 ,0xdc72,0xcac9 ,
+  0xc44a,0x1709 ,0x0bb7,0x3eeb ,0xde6d,0xc983 ,
+  0xc3be,0x1590 ,0x0af1,0x3f0f ,0xe074,0xc850 ,
+  0xc33b,0x1413 ,0x0a2b,0x3f30 ,0xe287,0xc731 ,
+  0xc2c1,0x1294 ,0x0964,0x3f4f ,0xe4a3,0xc625 ,
+  0xc251,0x1112 ,0x089d,0x3f6b ,0xe6c9,0xc52d ,
+  0xc1eb,0x0f8d ,0x07d6,0x3f85 ,0xe8f7,0xc44a ,
+  0xc18e,0x0e06 ,0x070e,0x3f9c ,0xeb2e,0xc37b ,
+  0xc13b,0x0c7c ,0x0646,0x3fb1 ,0xed6c,0xc2c1 ,
+  0xc0f1,0x0af1 ,0x057e,0x3fc4 ,0xefb0,0xc21d ,
+  0xc0b1,0x0964 ,0x04b5,0x3fd4 ,0xf1fa,0xc18e ,
+  0xc07b,0x07d6 ,0x03ed,0x3fe1 ,0xf449,0xc115 ,
+  0xc04f,0x0646 ,0x0324,0x3fec ,0xf69c,0xc0b1 ,
+  0xc02c,0x04b5 ,0x025b,0x3ff5 ,0xf8f2,0xc064 ,
+  0xc014,0x0324 ,0x0192,0x3ffb ,0xfb4b,0xc02c ,
+  0xc005,0x0192 ,0x00c9,0x3fff ,0xfda5,0xc00b ,
+  0x4000,0x0000 ,0x4000,0x0065 ,0x3fff,0x00c9 ,
+  0x3ffd,0x012e ,0x3ffb,0x0192 ,0x3ff8,0x01f7 ,
+  0x3ff5,0x025b ,0x3ff1,0x02c0 ,0x3fec,0x0324 ,
+  0x3fe7,0x0388 ,0x3fe1,0x03ed ,0x3fdb,0x0451 ,
+  0x3fd4,0x04b5 ,0x3fcc,0x051a ,0x3fc4,0x057e ,
+  0x3fbb,0x05e2 ,0x3fb1,0x0646 ,0x3fa7,0x06aa ,
+  0x3f9c,0x070e ,0x3f91,0x0772 ,0x3f85,0x07d6 ,
+  0x3f78,0x0839 ,0x3f6b,0x089d ,0x3f5d,0x0901 ,
+  0x3f4f,0x0964 ,0x3f40,0x09c7 ,0x3f30,0x0a2b ,
+  0x3f20,0x0a8e ,0x3f0f,0x0af1 ,0x3efd,0x0b54 ,
+  0x3eeb,0x0bb7 ,0x3ed8,0x0c1a ,0x3ec5,0x0c7c ,
+  0x3eb1,0x0cdf ,0x3e9d,0x0d41 ,0x3e88,0x0da4 ,
+  0x3e72,0x0e06 ,0x3e5c,0x0e68 ,0x3e45,0x0eca ,
+  0x3e2d,0x0f2b ,0x3e15,0x0f8d ,0x3dfc,0x0fee ,
+  0x3de3,0x1050 ,0x3dc9,0x10b1 ,0x3daf,0x1112 ,
+  0x3d93,0x1173 ,0x3d78,0x11d3 ,0x3d5b,0x1234 ,
+  0x3d3f,0x1294 ,0x3d21,0x12f4 ,0x3d03,0x1354 ,
+  0x3ce4,0x13b4 ,0x3cc5,0x1413 ,0x3ca5,0x1473 ,
+  0x3c85,0x14d2 ,0x3c64,0x1531 ,0x3c42,0x1590 ,
+  0x3c20,0x15ee ,0x3bfd,0x164c ,0x3bda,0x16ab ,
+  0x3bb6,0x1709 ,0x3b92,0x1766 ,0x3b6d,0x17c4 ,
+  0x3b47,0x1821 ,0x3b21,0x187e ,0x3afa,0x18db ,
+  0x3ad3,0x1937 ,0x3aab,0x1993 ,0x3a82,0x19ef ,
+  0x3a59,0x1a4b ,0x3a30,0x1aa7 ,0x3a06,0x1b02 ,
+  0x39db,0x1b5d ,0x39b0,0x1bb8 ,0x3984,0x1c12 ,
+  0x3958,0x1c6c ,0x392b,0x1cc6 ,0x38fd,0x1d20 ,
+  0x38cf,0x1d79 ,0x38a1,0x1dd3 ,0x3871,0x1e2b ,
+  0x3842,0x1e84 ,0x3812,0x1edc ,0x37e1,0x1f34 ,
+  0x37b0,0x1f8c ,0x377e,0x1fe3 ,0x374b,0x203a ,
+  0x3718,0x2091 ,0x36e5,0x20e7 ,0x36b1,0x213d ,
+  0x367d,0x2193 ,0x3648,0x21e8 ,0x3612,0x223d ,
+  0x35dc,0x2292 ,0x35a5,0x22e7 ,0x356e,0x233b ,
+  0x3537,0x238e ,0x34ff,0x23e2 ,0x34c6,0x2435 ,
+  0x348d,0x2488 ,0x3453,0x24da ,0x3419,0x252c ,
+  0x33df,0x257e ,0x33a3,0x25cf ,0x3368,0x2620 ,
+  0x332c,0x2671 ,0x32ef,0x26c1 ,0x32b2,0x2711 ,
+  0x3274,0x2760 ,0x3236,0x27af ,0x31f8,0x27fe ,
+  0x31b9,0x284c ,0x3179,0x289a ,0x3139,0x28e7 ,
+  0x30f9,0x2935 ,0x30b8,0x2981 ,0x3076,0x29ce ,
+  0x3034,0x2a1a ,0x2ff2,0x2a65 ,0x2faf,0x2ab0 ,
+  0x2f6c,0x2afb ,0x2f28,0x2b45 ,0x2ee4,0x2b8f ,
+  0x2e9f,0x2bd8 ,0x2e5a,0x2c21 ,0x2e15,0x2c6a ,
+  0x2dcf,0x2cb2 ,0x2d88,0x2cfa ,0x2d41,0x2d41 ,
+  0x2cfa,0x2d88 ,0x2cb2,0x2dcf ,0x2c6a,0x2e15 ,
+  0x2c21,0x2e5a ,0x2bd8,0x2e9f ,0x2b8f,0x2ee4 ,
+  0x2b45,0x2f28 ,0x2afb,0x2f6c ,0x2ab0,0x2faf ,
+  0x2a65,0x2ff2 ,0x2a1a,0x3034 ,0x29ce,0x3076 ,
+  0x2981,0x30b8 ,0x2935,0x30f9 ,0x28e7,0x3139 ,
+  0x289a,0x3179 ,0x284c,0x31b9 ,0x27fe,0x31f8 ,
+  0x27af,0x3236 ,0x2760,0x3274 ,0x2711,0x32b2 ,
+  0x26c1,0x32ef ,0x2671,0x332c ,0x2620,0x3368 ,
+  0x25cf,0x33a3 ,0x257e,0x33df ,0x252c,0x3419 ,
+  0x24da,0x3453 ,0x2488,0x348d ,0x2435,0x34c6 ,
+  0x23e2,0x34ff ,0x238e,0x3537 ,0x233b,0x356e ,
+  0x22e7,0x35a5 ,0x2292,0x35dc ,0x223d,0x3612 ,
+  0x21e8,0x3648 ,0x2193,0x367d ,0x213d,0x36b1 ,
+  0x20e7,0x36e5 ,0x2091,0x3718 ,0x203a,0x374b ,
+  0x1fe3,0x377e ,0x1f8c,0x37b0 ,0x1f34,0x37e1 ,
+  0x1edc,0x3812 ,0x1e84,0x3842 ,0x1e2b,0x3871 ,
+  0x1dd3,0x38a1 ,0x1d79,0x38cf ,0x1d20,0x38fd ,
+  0x1cc6,0x392b ,0x1c6c,0x3958 ,0x1c12,0x3984 ,
+  0x1bb8,0x39b0 ,0x1b5d,0x39db ,0x1b02,0x3a06 ,
+  0x1aa7,0x3a30 ,0x1a4b,0x3a59 ,0x19ef,0x3a82 ,
+  0x1993,0x3aab ,0x1937,0x3ad3 ,0x18db,0x3afa ,
+  0x187e,0x3b21 ,0x1821,0x3b47 ,0x17c4,0x3b6d ,
+  0x1766,0x3b92 ,0x1709,0x3bb6 ,0x16ab,0x3bda ,
+  0x164c,0x3bfd ,0x15ee,0x3c20 ,0x1590,0x3c42 ,
+  0x1531,0x3c64 ,0x14d2,0x3c85 ,0x1473,0x3ca5 ,
+  0x1413,0x3cc5 ,0x13b4,0x3ce4 ,0x1354,0x3d03 ,
+  0x12f4,0x3d21 ,0x1294,0x3d3f ,0x1234,0x3d5b ,
+  0x11d3,0x3d78 ,0x1173,0x3d93 ,0x1112,0x3daf ,
+  0x10b1,0x3dc9 ,0x1050,0x3de3 ,0x0fee,0x3dfc ,
+  0x0f8d,0x3e15 ,0x0f2b,0x3e2d ,0x0eca,0x3e45 ,
+  0x0e68,0x3e5c ,0x0e06,0x3e72 ,0x0da4,0x3e88 ,
+  0x0d41,0x3e9d ,0x0cdf,0x3eb1 ,0x0c7c,0x3ec5 ,
+  0x0c1a,0x3ed8 ,0x0bb7,0x3eeb ,0x0b54,0x3efd ,
+  0x0af1,0x3f0f ,0x0a8e,0x3f20 ,0x0a2b,0x3f30 ,
+  0x09c7,0x3f40 ,0x0964,0x3f4f ,0x0901,0x3f5d ,
+  0x089d,0x3f6b ,0x0839,0x3f78 ,0x07d6,0x3f85 ,
+  0x0772,0x3f91 ,0x070e,0x3f9c ,0x06aa,0x3fa7 ,
+  0x0646,0x3fb1 ,0x05e2,0x3fbb ,0x057e,0x3fc4 ,
+  0x051a,0x3fcc ,0x04b5,0x3fd4 ,0x0451,0x3fdb ,
+  0x03ed,0x3fe1 ,0x0388,0x3fe7 ,0x0324,0x3fec ,
+  0x02c0,0x3ff1 ,0x025b,0x3ff5 ,0x01f7,0x3ff8 ,
+  0x0192,0x3ffb ,0x012e,0x3ffd ,0x00c9,0x3fff ,
+  0x0065,0x4000 ,0x0000,0x4000 ,0xff9b,0x4000 ,
+  0xff37,0x3fff ,0xfed2,0x3ffd ,0xfe6e,0x3ffb ,
+  0xfe09,0x3ff8 ,0xfda5,0x3ff5 ,0xfd40,0x3ff1 ,
+  0xfcdc,0x3fec ,0xfc78,0x3fe7 ,0xfc13,0x3fe1 ,
+  0xfbaf,0x3fdb ,0xfb4b,0x3fd4 ,0xfae6,0x3fcc ,
+  0xfa82,0x3fc4 ,0xfa1e,0x3fbb ,0xf9ba,0x3fb1 ,
+  0xf956,0x3fa7 ,0xf8f2,0x3f9c ,0xf88e,0x3f91 ,
+  0xf82a,0x3f85 ,0xf7c7,0x3f78 ,0xf763,0x3f6b ,
+  0xf6ff,0x3f5d ,0xf69c,0x3f4f ,0xf639,0x3f40 ,
+  0xf5d5,0x3f30 ,0xf572,0x3f20 ,0xf50f,0x3f0f ,
+  0xf4ac,0x3efd ,0xf449,0x3eeb ,0xf3e6,0x3ed8 ,
+  0xf384,0x3ec5 ,0xf321,0x3eb1 ,0xf2bf,0x3e9d ,
+  0xf25c,0x3e88 ,0xf1fa,0x3e72 ,0xf198,0x3e5c ,
+  0xf136,0x3e45 ,0xf0d5,0x3e2d ,0xf073,0x3e15 ,
+  0xf012,0x3dfc ,0xefb0,0x3de3 ,0xef4f,0x3dc9 ,
+  0xeeee,0x3daf ,0xee8d,0x3d93 ,0xee2d,0x3d78 ,
+  0xedcc,0x3d5b ,0xed6c,0x3d3f ,0xed0c,0x3d21 ,
+  0xecac,0x3d03 ,0xec4c,0x3ce4 ,0xebed,0x3cc5 ,
+  0xeb8d,0x3ca5 ,0xeb2e,0x3c85 ,0xeacf,0x3c64 ,
+  0xea70,0x3c42 ,0xea12,0x3c20 ,0xe9b4,0x3bfd ,
+  0xe955,0x3bda ,0xe8f7,0x3bb6 ,0xe89a,0x3b92 ,
+  0xe83c,0x3b6d ,0xe7df,0x3b47 ,0xe782,0x3b21 ,
+  0xe725,0x3afa ,0xe6c9,0x3ad3 ,0xe66d,0x3aab ,
+  0xe611,0x3a82 ,0xe5b5,0x3a59 ,0xe559,0x3a30 ,
+  0xe4fe,0x3a06 ,0xe4a3,0x39db ,0xe448,0x39b0 ,
+  0xe3ee,0x3984 ,0xe394,0x3958 ,0xe33a,0x392b ,
+  0xe2e0,0x38fd ,0xe287,0x38cf ,0xe22d,0x38a1 ,
+  0xe1d5,0x3871 ,0xe17c,0x3842 ,0xe124,0x3812 ,
+  0xe0cc,0x37e1 ,0xe074,0x37b0 ,0xe01d,0x377e ,
+  0xdfc6,0x374b ,0xdf6f,0x3718 ,0xdf19,0x36e5 ,
+  0xdec3,0x36b1 ,0xde6d,0x367d ,0xde18,0x3648 ,
+  0xddc3,0x3612 ,0xdd6e,0x35dc ,0xdd19,0x35a5 ,
+  0xdcc5,0x356e ,0xdc72,0x3537 ,0xdc1e,0x34ff ,
+  0xdbcb,0x34c6 ,0xdb78,0x348d ,0xdb26,0x3453 ,
+  0xdad4,0x3419 ,0xda82,0x33df ,0xda31,0x33a3 ,
+  0xd9e0,0x3368 ,0xd98f,0x332c ,0xd93f,0x32ef ,
+  0xd8ef,0x32b2 ,0xd8a0,0x3274 ,0xd851,0x3236 ,
+  0xd802,0x31f8 ,0xd7b4,0x31b9 ,0xd766,0x3179 ,
+  0xd719,0x3139 ,0xd6cb,0x30f9 ,0xd67f,0x30b8 ,
+  0xd632,0x3076 ,0xd5e6,0x3034 ,0xd59b,0x2ff2 ,
+  0xd550,0x2faf ,0xd505,0x2f6c ,0xd4bb,0x2f28 ,
+  0xd471,0x2ee4 ,0xd428,0x2e9f ,0xd3df,0x2e5a ,
+  0xd396,0x2e15 ,0xd34e,0x2dcf ,0xd306,0x2d88 ,
+  0xd2bf,0x2d41 ,0xd278,0x2cfa ,0xd231,0x2cb2 ,
+  0xd1eb,0x2c6a ,0xd1a6,0x2c21 ,0xd161,0x2bd8 ,
+  0xd11c,0x2b8f ,0xd0d8,0x2b45 ,0xd094,0x2afb ,
+  0xd051,0x2ab0 ,0xd00e,0x2a65 ,0xcfcc,0x2a1a ,
+  0xcf8a,0x29ce ,0xcf48,0x2981 ,0xcf07,0x2935 ,
+  0xcec7,0x28e7 ,0xce87,0x289a ,0xce47,0x284c ,
+  0xce08,0x27fe ,0xcdca,0x27af ,0xcd8c,0x2760 ,
+  0xcd4e,0x2711 ,0xcd11,0x26c1 ,0xccd4,0x2671 ,
+  0xcc98,0x2620 ,0xcc5d,0x25cf ,0xcc21,0x257e ,
+  0xcbe7,0x252c ,0xcbad,0x24da ,0xcb73,0x2488 ,
+  0xcb3a,0x2435 ,0xcb01,0x23e2 ,0xcac9,0x238e ,
+  0xca92,0x233b ,0xca5b,0x22e7 ,0xca24,0x2292 ,
+  0xc9ee,0x223d ,0xc9b8,0x21e8 ,0xc983,0x2193 ,
+  0xc94f,0x213d ,0xc91b,0x20e7 ,0xc8e8,0x2091 ,
+  0xc8b5,0x203a ,0xc882,0x1fe3 ,0xc850,0x1f8c ,
+  0xc81f,0x1f34 ,0xc7ee,0x1edc ,0xc7be,0x1e84 ,
+  0xc78f,0x1e2b ,0xc75f,0x1dd3 ,0xc731,0x1d79 ,
+  0xc703,0x1d20 ,0xc6d5,0x1cc6 ,0xc6a8,0x1c6c ,
+  0xc67c,0x1c12 ,0xc650,0x1bb8 ,0xc625,0x1b5d ,
+  0xc5fa,0x1b02 ,0xc5d0,0x1aa7 ,0xc5a7,0x1a4b ,
+  0xc57e,0x19ef ,0xc555,0x1993 ,0xc52d,0x1937 ,
+  0xc506,0x18db ,0xc4df,0x187e ,0xc4b9,0x1821 ,
+  0xc493,0x17c4 ,0xc46e,0x1766 ,0xc44a,0x1709 ,
+  0xc426,0x16ab ,0xc403,0x164c ,0xc3e0,0x15ee ,
+  0xc3be,0x1590 ,0xc39c,0x1531 ,0xc37b,0x14d2 ,
+  0xc35b,0x1473 ,0xc33b,0x1413 ,0xc31c,0x13b4 ,
+  0xc2fd,0x1354 ,0xc2df,0x12f4 ,0xc2c1,0x1294 ,
+  0xc2a5,0x1234 ,0xc288,0x11d3 ,0xc26d,0x1173 ,
+  0xc251,0x1112 ,0xc237,0x10b1 ,0xc21d,0x1050 ,
+  0xc204,0x0fee ,0xc1eb,0x0f8d ,0xc1d3,0x0f2b ,
+  0xc1bb,0x0eca ,0xc1a4,0x0e68 ,0xc18e,0x0e06 ,
+  0xc178,0x0da4 ,0xc163,0x0d41 ,0xc14f,0x0cdf ,
+  0xc13b,0x0c7c ,0xc128,0x0c1a ,0xc115,0x0bb7 ,
+  0xc103,0x0b54 ,0xc0f1,0x0af1 ,0xc0e0,0x0a8e ,
+  0xc0d0,0x0a2b ,0xc0c0,0x09c7 ,0xc0b1,0x0964 ,
+  0xc0a3,0x0901 ,0xc095,0x089d ,0xc088,0x0839 ,
+  0xc07b,0x07d6 ,0xc06f,0x0772 ,0xc064,0x070e ,
+  0xc059,0x06aa ,0xc04f,0x0646 ,0xc045,0x05e2 ,
+  0xc03c,0x057e ,0xc034,0x051a ,0xc02c,0x04b5 ,
+  0xc025,0x0451 ,0xc01f,0x03ed ,0xc019,0x0388 ,
+  0xc014,0x0324 ,0xc00f,0x02c0 ,0xc00b,0x025b ,
+  0xc008,0x01f7 ,0xc005,0x0192 ,0xc003,0x012e ,
+  0xc001,0x00c9 ,0xc000,0x0065 };

src/common_audio/signal_processing_library/main/source/webrtc_fft_t_rad.c

@@ -0,0 +1,27 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the Q14 radix-2 tables used in ARM9E optimization routines.
+ *
+ */
+
+extern const unsigned short t_Q14S_rad8[2];
+const unsigned short t_Q14S_rad8[2] = {  0x0000,0x2d41 };
+
+//extern const int t_Q30S_rad8[2];
+//const int t_Q30S_rad8[2] = {  0x00000000,0x2d413ccd };
+
+extern const unsigned short t_Q14R_rad8[2];
+const unsigned short t_Q14R_rad8[2] = {  0x2d41,0x2d41 };
+
+//extern const int t_Q30R_rad8[2];
+//const int t_Q30R_rad8[2] = {  0x2d413ccd,0x2d413ccd };

src/common_audio/signal_processing_library/main/test/unit_test/unit_test.cc

@@ -0,0 +1,494 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file contains the SPL unit_test.
+ *
+ */
+
+#include "unit_test.h"
+#include "signal_processing_library.h"
+
+class SplEnvironment : public ::testing::Environment {
+ public:
+  virtual void SetUp() {
+  }
+  virtual void TearDown() {
+  }
+};
+
+SplTest::SplTest()
+{
+}
+
+void SplTest::SetUp() {
+}
+
+void SplTest::TearDown() {
+}
+
+TEST_F(SplTest, MacroTest) {
+    // Macros with inputs.
+    int A = 10;
+    int B = 21;
+    int a = -3;
+    int b = WEBRTC_SPL_WORD32_MAX;
+    int nr = 2;
+    int d_ptr1 = 0;
+    int d_ptr2 = 0;
+
+    EXPECT_EQ(10, WEBRTC_SPL_MIN(A, B));
+    EXPECT_EQ(21, WEBRTC_SPL_MAX(A, B));
+
+    EXPECT_EQ(3, WEBRTC_SPL_ABS_W16(a));
+    EXPECT_EQ(3, WEBRTC_SPL_ABS_W32(a));
+    EXPECT_EQ(0, WEBRTC_SPL_GET_BYTE(&B, nr));
+    WEBRTC_SPL_SET_BYTE(&d_ptr2, 1, nr);
+    EXPECT_EQ(65536, d_ptr2);
+
+    EXPECT_EQ(-63, WEBRTC_SPL_MUL(a, B));
+    EXPECT_EQ(-2147483645, WEBRTC_SPL_MUL(a, b));
+    EXPECT_EQ(-2147483645, WEBRTC_SPL_UMUL(a, b));
+    b = WEBRTC_SPL_WORD16_MAX >> 1;
+    EXPECT_EQ(65535, WEBRTC_SPL_UMUL_RSFT16(a, b));
+    EXPECT_EQ(1073627139, WEBRTC_SPL_UMUL_16_16(a, b));
+    EXPECT_EQ(16382, WEBRTC_SPL_UMUL_16_16_RSFT16(a, b));
+    EXPECT_EQ(-49149, WEBRTC_SPL_UMUL_32_16(a, b));
+    EXPECT_EQ(65535, WEBRTC_SPL_UMUL_32_16_RSFT16(a, b));
+    EXPECT_EQ(-49149, WEBRTC_SPL_MUL_16_U16(a, b));
+
+    a = b;
+    b = -3;
+    EXPECT_EQ(-5461, WEBRTC_SPL_DIV(a, b));
+    EXPECT_EQ(0, WEBRTC_SPL_UDIV(a, b));
+
+    EXPECT_EQ(-1, WEBRTC_SPL_MUL_16_32_RSFT16(a, b));
+    EXPECT_EQ(-1, WEBRTC_SPL_MUL_16_32_RSFT15(a, b));
+    EXPECT_EQ(-3, WEBRTC_SPL_MUL_16_32_RSFT14(a, b));
+    EXPECT_EQ(-24, WEBRTC_SPL_MUL_16_32_RSFT11(a, b));
+
+    int a32 = WEBRTC_SPL_WORD32_MAX;
+    int a32a = (WEBRTC_SPL_WORD32_MAX >> 16);
+    int a32b = (WEBRTC_SPL_WORD32_MAX & 0x0000ffff);
+    EXPECT_EQ(5, WEBRTC_SPL_MUL_32_32_RSFT32(a32a, a32b, A));
+    EXPECT_EQ(5, WEBRTC_SPL_MUL_32_32_RSFT32BI(a32, A));
+
+    EXPECT_EQ(-49149, WEBRTC_SPL_MUL_16_16(a, b));
+    EXPECT_EQ(-12288, WEBRTC_SPL_MUL_16_16_RSFT(a, b, 2));
+
+    EXPECT_EQ(-12287, WEBRTC_SPL_MUL_16_16_RSFT_WITH_ROUND(a, b, 2));
+    EXPECT_EQ(-1, WEBRTC_SPL_MUL_16_16_RSFT_WITH_FIXROUND(a, b));
+
+    EXPECT_EQ(16380, WEBRTC_SPL_ADD_SAT_W32(a, b));
+    EXPECT_EQ(21, WEBRTC_SPL_SAT(a, A, B));
+    EXPECT_EQ(21, WEBRTC_SPL_SAT(a, B, A));
+    EXPECT_EQ(-49149, WEBRTC_SPL_MUL_32_16(a, b));
+
+    EXPECT_EQ(16386, WEBRTC_SPL_SUB_SAT_W32(a, b));
+    EXPECT_EQ(16380, WEBRTC_SPL_ADD_SAT_W16(a, b));
+    EXPECT_EQ(16386, WEBRTC_SPL_SUB_SAT_W16(a, b));
+
+    EXPECT_TRUE(WEBRTC_SPL_IS_NEG(b));
+
+    // Shifting with negative numbers allowed
+    // Positive means left shift
+    EXPECT_EQ(32766, WEBRTC_SPL_SHIFT_W16(a, 1));
+    EXPECT_EQ(32766, WEBRTC_SPL_SHIFT_W32(a, 1));
+
+    // Shifting with negative numbers not allowed
+    // We cannot do casting here due to signed/unsigned problem
+    EXPECT_EQ(8191, WEBRTC_SPL_RSHIFT_W16(a, 1));
+    EXPECT_EQ(32766, WEBRTC_SPL_LSHIFT_W16(a, 1));
+    EXPECT_EQ(8191, WEBRTC_SPL_RSHIFT_W32(a, 1));
+    EXPECT_EQ(32766, WEBRTC_SPL_LSHIFT_W32(a, 1));
+
+    EXPECT_EQ(8191, WEBRTC_SPL_RSHIFT_U16(a, 1));
+    EXPECT_EQ(32766, WEBRTC_SPL_LSHIFT_U16(a, 1));
+    EXPECT_EQ(8191, WEBRTC_SPL_RSHIFT_U32(a, 1));
+    EXPECT_EQ(32766, WEBRTC_SPL_LSHIFT_U32(a, 1));
+
+    EXPECT_EQ(1470, WEBRTC_SPL_RAND(A));
+}
+
+TEST_F(SplTest, InlineTest) {
+
+    WebRtc_Word16 a = 121;
+    WebRtc_Word16 b = -17;
+    WebRtc_Word32 A = 111121;
+    WebRtc_Word32 B = -1711;
+    char bVersion[8];
+
+    EXPECT_EQ(104, WebRtcSpl_AddSatW16(a, b));
+    EXPECT_EQ(138, WebRtcSpl_SubSatW16(a, b));
+
+    EXPECT_EQ(109410, WebRtcSpl_AddSatW32(A, B));
+    EXPECT_EQ(112832, WebRtcSpl_SubSatW32(A, B));
+
+    EXPECT_EQ(17, WebRtcSpl_GetSizeInBits(A));
+    EXPECT_EQ(14, WebRtcSpl_NormW32(A));
+    EXPECT_EQ(4, WebRtcSpl_NormW16(B));
+    EXPECT_EQ(15, WebRtcSpl_NormU32(A));
+
+    EXPECT_EQ(0, WebRtcSpl_get_version(bVersion, 8));
+}
+
+TEST_F(SplTest, MathOperationsTest) {
+
+    int A = 117;
+    WebRtc_Word32 num = 117;
+    WebRtc_Word32 den = -5;
+    WebRtc_UWord16 denU = 5;
+    EXPECT_EQ(10, WebRtcSpl_Sqrt(A));
+    EXPECT_EQ(10, WebRtcSpl_SqrtFloor(A));
+
+
+    EXPECT_EQ(-91772805, WebRtcSpl_DivResultInQ31(den, num));
+    EXPECT_EQ(-23, WebRtcSpl_DivW32W16ResW16(num, (WebRtc_Word16)den));
+    EXPECT_EQ(-23, WebRtcSpl_DivW32W16(num, (WebRtc_Word16)den));
+    EXPECT_EQ(23, WebRtcSpl_DivU32U16(num, denU));
+    EXPECT_EQ(0, WebRtcSpl_DivW32HiLow(128, 0, 256));
+}
+
+TEST_F(SplTest, BasicArrayOperationsTest) {
+
+
+    const int kVectorSize = 4;
+    int B[] = {4, 12, 133, 1100};
+    int Bs[] = {2, 6, 66, 550};
+    WebRtc_UWord8 b8[kVectorSize];
+    WebRtc_Word16 b16[kVectorSize];
+    WebRtc_Word32 b32[kVectorSize];
+
+    WebRtc_UWord8 bTmp8[kVectorSize];
+    WebRtc_Word16 bTmp16[kVectorSize];
+    WebRtc_Word32 bTmp32[kVectorSize];
+
+    WebRtcSpl_MemSetW16(b16, 3, kVectorSize);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ(3, b16[kk]);
+    }
+    EXPECT_EQ(kVectorSize, WebRtcSpl_ZerosArrayW16(b16, kVectorSize));
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ(0, b16[kk]);
+    }
+    EXPECT_EQ(kVectorSize, WebRtcSpl_OnesArrayW16(b16, kVectorSize));
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ(1, b16[kk]);
+    }
+    WebRtcSpl_MemSetW32(b32, 3, kVectorSize);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ(3, b32[kk]);
+    }
+    EXPECT_EQ(kVectorSize, WebRtcSpl_ZerosArrayW32(b32, kVectorSize));
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ(0, b32[kk]);
+    }
+    EXPECT_EQ(kVectorSize, WebRtcSpl_OnesArrayW32(b32, kVectorSize));
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ(1, b32[kk]);
+    }
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        bTmp8[kk] = (WebRtc_Word8)kk;
+        bTmp16[kk] = (WebRtc_Word16)kk;
+        bTmp32[kk] = (WebRtc_Word32)kk;
+    }
+    WEBRTC_SPL_MEMCPY_W8(b8, bTmp8, kVectorSize);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ(b8[kk], bTmp8[kk]);
+    }
+    WEBRTC_SPL_MEMCPY_W16(b16, bTmp16, kVectorSize);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ(b16[kk], bTmp16[kk]);
+    }
+//    WEBRTC_SPL_MEMCPY_W32(b32, bTmp32, kVectorSize);
+//    for (int kk = 0; kk < kVectorSize; ++kk) {
+//        EXPECT_EQ(b32[kk], bTmp32[kk]);
+//    }
+    EXPECT_EQ(2, WebRtcSpl_CopyFromEndW16(b16, kVectorSize, 2, bTmp16));
+    for (int kk = 0; kk < 2; ++kk) {
+        EXPECT_EQ(kk+2, bTmp16[kk]);
+    }
+
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        b32[kk] = B[kk];
+        b16[kk] = (WebRtc_Word16)B[kk];
+    }
+    WebRtcSpl_VectorBitShiftW32ToW16(bTmp16, kVectorSize, b32, 1);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ((B[kk]>>1), bTmp16[kk]);
+    }
+    WebRtcSpl_VectorBitShiftW16(bTmp16, kVectorSize, b16, 1);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ((B[kk]>>1), bTmp16[kk]);
+    }
+    WebRtcSpl_VectorBitShiftW32(bTmp32, kVectorSize, b32, 1);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ((B[kk]>>1), bTmp32[kk]);
+    }
+
+    WebRtcSpl_MemCpyReversedOrder(&bTmp16[3], b16, kVectorSize);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ(b16[3-kk], bTmp16[kk]);
+    }
+
+}
+
+TEST_F(SplTest, MinMaxOperationsTest) {
+
+
+    const int kVectorSize = 4;
+    int B[] = {4, 12, 133, -1100};
+    WebRtc_Word16 b16[kVectorSize];
+    WebRtc_Word32 b32[kVectorSize];
+
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        b16[kk] = B[kk];
+        b32[kk] = B[kk];
+    }
+
+    EXPECT_EQ(1100, WebRtcSpl_MaxAbsValueW16(b16, kVectorSize));
+    EXPECT_EQ(1100, WebRtcSpl_MaxAbsValueW32(b32, kVectorSize));
+    EXPECT_EQ(133, WebRtcSpl_MaxValueW16(b16, kVectorSize));
+    EXPECT_EQ(133, WebRtcSpl_MaxValueW32(b32, kVectorSize));
+    EXPECT_EQ(3, WebRtcSpl_MaxAbsIndexW16(b16, kVectorSize));
+    EXPECT_EQ(2, WebRtcSpl_MaxIndexW16(b16, kVectorSize));
+    EXPECT_EQ(2, WebRtcSpl_MaxIndexW32(b32, kVectorSize));
+
+    EXPECT_EQ(-1100, WebRtcSpl_MinValueW16(b16, kVectorSize));
+    EXPECT_EQ(-1100, WebRtcSpl_MinValueW32(b32, kVectorSize));
+    EXPECT_EQ(3, WebRtcSpl_MinIndexW16(b16, kVectorSize));
+    EXPECT_EQ(3, WebRtcSpl_MinIndexW32(b32, kVectorSize));
+
+    EXPECT_EQ(0, WebRtcSpl_GetScalingSquare(b16, kVectorSize, 1));
+
+}
+
+TEST_F(SplTest, VectorOperationsTest) {
+
+
+    const int kVectorSize = 4;
+    int B[] = {4, 12, 133, 1100};
+    WebRtc_Word16 a16[kVectorSize];
+    WebRtc_Word16 b16[kVectorSize];
+    WebRtc_Word32 b32[kVectorSize];
+    WebRtc_Word16 bTmp16[kVectorSize];
+
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        a16[kk] = B[kk];
+        b16[kk] = B[kk];
+    }
+
+    WebRtcSpl_AffineTransformVector(bTmp16, b16, 3, 7, 2, kVectorSize);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ((B[kk]*3+7)>>2, bTmp16[kk]);
+    }
+    WebRtcSpl_ScaleAndAddVectorsWithRound(b16, 3, b16, 2, 2, bTmp16, kVectorSize);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ((B[kk]*3+B[kk]*2+2)>>2, bTmp16[kk]);
+    }
+
+    WebRtcSpl_AddAffineVectorToVector(bTmp16, b16, 3, 7, 2, kVectorSize);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ(((B[kk]*3+B[kk]*2+2)>>2)+((b16[kk]*3+7)>>2), bTmp16[kk]);
+    }
+
+    WebRtcSpl_CrossCorrelation(b32, b16, bTmp16, kVectorSize, 2, 2, 0);
+    for (int kk = 0; kk < 2; ++kk) {
+        EXPECT_EQ(614236, b32[kk]);
+    }
+//    EXPECT_EQ(, WebRtcSpl_DotProduct(b16, bTmp16, 4));
+    EXPECT_EQ(306962, WebRtcSpl_DotProductWithScale(b16, b16, kVectorSize, 2));
+
+    WebRtcSpl_ScaleVector(b16, bTmp16, 13, kVectorSize, 2);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ((b16[kk]*13)>>2, bTmp16[kk]);
+    }
+    WebRtcSpl_ScaleVectorWithSat(b16, bTmp16, 13, kVectorSize, 2);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ((b16[kk]*13)>>2, bTmp16[kk]);
+    }
+    WebRtcSpl_ScaleAndAddVectors(a16, 13, 2, b16, 7, 2, bTmp16, kVectorSize);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ(((a16[kk]*13)>>2)+((b16[kk]*7)>>2), bTmp16[kk]);
+    }
+
+    WebRtcSpl_AddVectorsAndShift(bTmp16, a16, b16, kVectorSize, 2);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ(B[kk] >> 1, bTmp16[kk]);
+    }
+    WebRtcSpl_ReverseOrderMultArrayElements(bTmp16, a16, &b16[3], kVectorSize, 2);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ((a16[kk]*b16[3-kk])>>2, bTmp16[kk]);
+    }
+    WebRtcSpl_ElementwiseVectorMult(bTmp16, a16, b16, kVectorSize, 6);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ((a16[kk]*b16[kk])>>6, bTmp16[kk]);
+    }
+
+    WebRtcSpl_SqrtOfOneMinusXSquared(b16, kVectorSize, bTmp16);
+    for (int kk = 0; kk < kVectorSize - 1; ++kk) {
+        EXPECT_EQ(32767, bTmp16[kk]);
+    }
+    EXPECT_EQ(32749, bTmp16[kVectorSize - 1]);
+}
+
+TEST_F(SplTest, EstimatorsTest) {
+
+
+    const int kVectorSize = 4;
+    int B[] = {4, 12, 133, 1100};
+    WebRtc_Word16 b16[kVectorSize];
+    WebRtc_Word32 b32[kVectorSize];
+    WebRtc_Word16 bTmp16[kVectorSize];
+
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        b16[kk] = B[kk];
+        b32[kk] = B[kk];
+    }
+
+    EXPECT_EQ(0, WebRtcSpl_LevinsonDurbin(b32, b16, bTmp16, 2));
+
+}
+
+TEST_F(SplTest, FilterTest) {
+
+
+    const int kVectorSize = 4;
+    WebRtc_Word16 A[] = {1, 2, 33, 100};
+    WebRtc_Word16 A5[] = {1, 2, 33, 100, -5};
+    WebRtc_Word16 B[] = {4, 12, 133, 110};
+    WebRtc_Word16 b16[kVectorSize];
+    WebRtc_Word16 bTmp16[kVectorSize];
+    WebRtc_Word16 bTmp16Low[kVectorSize];
+    WebRtc_Word16 bState[kVectorSize];
+    WebRtc_Word16 bStateLow[kVectorSize];
+
+    WebRtcSpl_ZerosArrayW16(bState, kVectorSize);
+    WebRtcSpl_ZerosArrayW16(bStateLow, kVectorSize);
+
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        b16[kk] = A[kk];
+    }
+
+    // MA filters
+    WebRtcSpl_FilterMAFastQ12(b16, bTmp16, B, kVectorSize, kVectorSize);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        //EXPECT_EQ(aTmp16[kk], bTmp16[kk]);
+    }
+    // AR filters
+    WebRtcSpl_FilterARFastQ12(b16, bTmp16, A, kVectorSize, kVectorSize);
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+//        EXPECT_EQ(aTmp16[kk], bTmp16[kk]);
+    }
+    EXPECT_EQ(kVectorSize, WebRtcSpl_FilterAR(A5,
+                                              5,
+                                              b16,
+                                              kVectorSize,
+                                              bState,
+                                              kVectorSize,
+                                              bStateLow,
+                                              kVectorSize,
+                                              bTmp16,
+                                              bTmp16Low,
+                                              kVectorSize));
+
+}
+
+TEST_F(SplTest, RandTest) {
+
+
+    const int kVectorSize = 4;
+    WebRtc_Word16 BU[] = {3653, 12446, 8525, 30691};
+    WebRtc_Word16 BN[] = {3459, -11689, -258, -3738};
+    WebRtc_Word16 b16[kVectorSize];
+    WebRtc_UWord32 bSeed = 100000;
+
+    EXPECT_EQ(464449057, WebRtcSpl_IncreaseSeed(&bSeed));
+    EXPECT_EQ(31565, WebRtcSpl_RandU(&bSeed));
+    EXPECT_EQ(-9786, WebRtcSpl_RandN(&bSeed));
+    EXPECT_EQ(kVectorSize, WebRtcSpl_RandUArray(b16, kVectorSize, &bSeed));
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        EXPECT_EQ(BU[kk], b16[kk]);
+    }
+}
+
+TEST_F(SplTest, SignalProcessingTest) {
+
+
+    const int kVectorSize = 4;
+    int A[] = {1, 2, 33, 100};
+    WebRtc_Word16 b16[kVectorSize];
+    WebRtc_Word32 b32[kVectorSize];
+
+    WebRtc_Word16 bTmp16[kVectorSize];
+    WebRtc_Word32 bTmp32[kVectorSize];
+
+    int bScale = 0;
+
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        b16[kk] = A[kk];
+        b32[kk] = A[kk];
+    }
+
+    EXPECT_EQ(2, WebRtcSpl_AutoCorrelation(b16, kVectorSize, 1, bTmp32, &bScale));
+    WebRtcSpl_ReflCoefToLpc(b16, kVectorSize, bTmp16);
+//    for (int kk = 0; kk < kVectorSize; ++kk) {
+//        EXPECT_EQ(aTmp16[kk], bTmp16[kk]);
+//    }
+    WebRtcSpl_LpcToReflCoef(bTmp16, kVectorSize, b16);
+//    for (int kk = 0; kk < kVectorSize; ++kk) {
+//        EXPECT_EQ(a16[kk], b16[kk]);
+//    }
+    WebRtcSpl_AutoCorrToReflCoef(b32, kVectorSize, bTmp16);
+//    for (int kk = 0; kk < kVectorSize; ++kk) {
+//        EXPECT_EQ(aTmp16[kk], bTmp16[kk]);
+//    }
+    WebRtcSpl_GetHanningWindow(bTmp16, kVectorSize);
+//    for (int kk = 0; kk < kVectorSize; ++kk) {
+//        EXPECT_EQ(aTmp16[kk], bTmp16[kk]);
+//    }
+
+    for (int kk = 0; kk < kVectorSize; ++kk) {
+        b16[kk] = A[kk];
+    }
+    EXPECT_EQ(11094 , WebRtcSpl_Energy(b16, kVectorSize, &bScale));
+    EXPECT_EQ(0, bScale);
+}
+
+TEST_F(SplTest, FFTTest) {
+
+
+    WebRtc_Word16 B[] = {1, 2, 33, 100,
+            2, 3, 34, 101,
+            3, 4, 35, 102,
+            4, 5, 36, 103};
+
+    EXPECT_EQ(0, WebRtcSpl_ComplexFFT(B, 3, 1));
+//    for (int kk = 0; kk < 16; ++kk) {
+//        EXPECT_EQ(A[kk], B[kk]);
+//    }
+    EXPECT_EQ(0, WebRtcSpl_ComplexIFFT(B, 3, 1));
+//    for (int kk = 0; kk < 16; ++kk) {
+//        EXPECT_EQ(A[kk], B[kk]);
+//    }
+    WebRtcSpl_ComplexBitReverse(B, 3);
+    for (int kk = 0; kk < 16; ++kk) {
+        //EXPECT_EQ(A[kk], B[kk]);
+    }
+}
+
+int main(int argc, char** argv) {
+  ::testing::InitGoogleTest(&argc, argv);
+  SplEnvironment* env = new SplEnvironment;
+  ::testing::AddGlobalTestEnvironment(env);
+
+  return RUN_ALL_TESTS();
+}

src/common_audio/signal_processing_library/main/test/unit_test/unit_test.h

@@ -0,0 +1,30 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+/*
+ * This header file contains the function WebRtcSpl_CopyFromBeginU8().
+ * The description header can be found in signal_processing_library.h
+ *
+ */
+
+#ifndef WEBRTC_SPL_UNIT_TEST_H_
+#define WEBRTC_SPL_UNIT_TEST_H_
+
+#include <gtest/gtest.h>
+
+class SplTest: public ::testing::Test
+{
+protected:
+    SplTest();
+    virtual void SetUp();
+    virtual void TearDown();
+};
+
+#endif  // WEBRTC_SPL_UNIT_TEST_H_

src/common_audio/vad/OWNERS

@@ -0,0 +1,2 @@
+bjornv@webrtc.org
+jan.skoglund@webrtc.org

src/common_audio/vad/main/interface/webrtc_vad.h

@@ -0,0 +1,159 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This header file includes the VAD API calls. Specific function calls are given below.
+ */
+
+#ifndef WEBRTC_VAD_WEBRTC_VAD_H_
+#define WEBRTC_VAD_WEBRTC_VAD_H_
+
+#include "typedefs.h"
+
+typedef struct WebRtcVadInst VadInst;
+
+#ifdef __cplusplus
+extern "C"
+{
+#endif
+
+/****************************************************************************
+ * WebRtcVad_get_version(...)
+ *
+ * This function returns the version number of the code.
+ *
+ * Output:
+ *      - version       : Pointer to a buffer where the version info will
+ *                        be stored.
+ * Input:
+ *      - size_bytes    : Size of the buffer.
+ *
+ */
+WebRtc_Word16 WebRtcVad_get_version(char *version, size_t size_bytes);
+
+/****************************************************************************
+ * WebRtcVad_AssignSize(...) 
+ *
+ * This functions get the size needed for storing the instance for encoder
+ * and decoder, respectively
+ *
+ * Input/Output:
+ *      - size_in_bytes : Pointer to integer where the size is returned
+ *
+ * Return value         : 0
+ */
+WebRtc_Word16 WebRtcVad_AssignSize(int *size_in_bytes);
+
+/****************************************************************************
+ * WebRtcVad_Assign(...) 
+ *
+ * This functions Assigns memory for the instances.
+ *
+ * Input:
+ *        - vad_inst_addr :  Address to where to assign memory
+ * Output:
+ *        - vad_inst      :  Pointer to the instance that should be created
+ *
+ * Return value           :  0 - Ok
+ *                          -1 - Error
+ */
+WebRtc_Word16 WebRtcVad_Assign(VadInst **vad_inst, void *vad_inst_addr);
+
+/****************************************************************************
+ * WebRtcVad_Create(...)
+ *
+ * This function creates an instance to the VAD structure
+ *
+ * Input:
+ *      - vad_inst      : Pointer to VAD instance that should be created
+ *
+ * Output:
+ *      - vad_inst      : Pointer to created VAD instance
+ *
+ * Return value         :  0 - Ok
+ *                        -1 - Error
+ */
+WebRtc_Word16 WebRtcVad_Create(VadInst **vad_inst);
+
+/****************************************************************************
+ * WebRtcVad_Free(...)
+ *
+ * This function frees the dynamic memory of a specified VAD instance
+ *
+ * Input:
+ *      - vad_inst      : Pointer to VAD instance that should be freed
+ *
+ * Return value         :  0 - Ok
+ *                        -1 - Error
+ */
+WebRtc_Word16 WebRtcVad_Free(VadInst *vad_inst);
+
+/****************************************************************************
+ * WebRtcVad_Init(...)
+ *
+ * This function initializes a VAD instance
+ *
+ * Input:
+ *      - vad_inst      : Instance that should be initialized
+ *
+ * Output:
+ *      - vad_inst      : Initialized instance
+ *
+ * Return value         :  0 - Ok
+ *                        -1 - Error
+ */
+WebRtc_Word16 WebRtcVad_Init(VadInst *vad_inst);
+
+/****************************************************************************
+ * WebRtcVad_set_mode(...)
+ *
+ * This function initializes a VAD instance
+ *
+ * Input:
+ *      - vad_inst      : VAD instance
+ *      - mode          : Aggressiveness setting (0, 1, 2, or 3) 
+ *
+ * Output:
+ *      - vad_inst      : Initialized instance
+ *
+ * Return value         :  0 - Ok
+ *                        -1 - Error
+ */
+WebRtc_Word16 WebRtcVad_set_mode(VadInst *vad_inst, WebRtc_Word16 mode);
+
+/****************************************************************************
+ * WebRtcVad_Process(...)
+ * 
+ * This functions does a VAD for the inserted speech frame
+ *
+ * Input
+ *        - vad_inst     : VAD Instance. Needs to be initiated before call.
+ *        - fs           : sampling frequency (Hz): 8000, 16000, or 32000
+ *        - speech_frame : Pointer to speech frame buffer
+ *        - frame_length : Length of speech frame buffer in number of samples
+ *
+ * Output:
+ *        - vad_inst     : Updated VAD instance
+ *
+ * Return value          :  1 - Active Voice
+ *                          0 - Non-active Voice
+ *                         -1 - Error
+ */
+WebRtc_Word16 WebRtcVad_Process(VadInst *vad_inst,
+                                WebRtc_Word16 fs,
+                                WebRtc_Word16 *speech_frame,
+                                WebRtc_Word16 frame_length);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif // WEBRTC_VAD_WEBRTC_VAD_H_

src/common_audio/vad/main/source/vad.gypi

@@ -0,0 +1,48 @@
+# Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+#
+# Use of this source code is governed by a BSD-style license
+# that can be found in the LICENSE file in the root of the source
+# tree. An additional intellectual property rights grant can be found
+# in the file PATENTS.  All contributing project authors may
+# be found in the AUTHORS file in the root of the source tree.
+
+{
+  'targets': [
+    {
+      'target_name': 'vad',
+      'type': '<(library)',
+      'dependencies': [
+        'spl',
+      ],
+      'include_dirs': [
+        '../interface',
+      ],
+      'direct_dependent_settings': {
+        'include_dirs': [
+          '../interface',
+        ],
+      },
+      'sources': [
+        '../interface/webrtc_vad.h',
+        'webrtc_vad.c',
+        'vad_const.c',
+        'vad_const.h',
+        'vad_defines.h',
+        'vad_core.c',
+        'vad_core.h',
+        'vad_filterbank.c',
+        'vad_filterbank.h',
+        'vad_gmm.c',
+        'vad_gmm.h',
+        'vad_sp.c',
+        'vad_sp.h',
+      ],
+    },
+  ],
+}
+
+# Local Variables:
+# tab-width:2
+# indent-tabs-mode:nil
+# End:
+# vim: set expandtab tabstop=2 shiftwidth=2:

src/common_audio/vad/main/source/vad_const.c

@@ -0,0 +1,80 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+/*
+ * This file includes the constant values used internally in VAD.
+ */
+
+#include "vad_const.h"
+
+// Spectrum Weighting
+const WebRtc_Word16 kSpectrumWeight[6] = {6, 8, 10, 12, 14, 16};
+
+const WebRtc_Word16 kCompVar = 22005;
+
+// Constant 160*log10(2) in Q9
+const WebRtc_Word16 kLogConst = 24660;
+
+// Constant log2(exp(1)) in Q12
+const WebRtc_Word16 kLog10Const = 5909;
+
+// Q15
+const WebRtc_Word16 kNoiseUpdateConst = 655;
+const WebRtc_Word16 kSpeechUpdateConst = 6554;
+
+// Q8
+const WebRtc_Word16 kBackEta = 154;
+
+// Coefficients used by WebRtcVad_HpOutput, Q14
+const WebRtc_Word16 kHpZeroCoefs[3] = {6631, -13262, 6631};
+const WebRtc_Word16 kHpPoleCoefs[3] = {16384, -7756, 5620};
+
+// Allpass filter coefficients, upper and lower, in Q15
+// Upper: 0.64, Lower: 0.17
+const WebRtc_Word16 kAllPassCoefsQ15[2] = {20972, 5571};
+const WebRtc_Word16 kAllPassCoefsQ13[2] = {5243, 1392}; // Q13
+
+// Minimum difference between the two models, Q5
+const WebRtc_Word16 kMinimumDifference[6] = {544, 544, 576, 576, 576, 576};
+
+// Upper limit of mean value for speech model, Q7
+const WebRtc_Word16 kMaximumSpeech[6] = {11392, 11392, 11520, 11520, 11520, 11520};
+
+// Minimum value for mean value
+const WebRtc_Word16 kMinimumMean[2] = {640, 768};
+
+// Upper limit of mean value for noise model, Q7
+const WebRtc_Word16 kMaximumNoise[6] = {9216, 9088, 8960, 8832, 8704, 8576};
+
+// Adjustment for division with two in WebRtcVad_SplitFilter
+const WebRtc_Word16 kOffsetVector[6] = {368, 368, 272, 176, 176, 176};
+
+// Start values for the Gaussian models, Q7
+// Weights for the two Gaussians for the six channels (noise)
+const WebRtc_Word16 kNoiseDataWeights[12] = {34, 62, 72, 66, 53, 25, 94, 66, 56, 62, 75, 103};
+
+// Weights for the two Gaussians for the six channels (speech)
+const WebRtc_Word16 kSpeechDataWeights[12] = {48, 82, 45, 87, 50, 47, 80, 46, 83, 41, 78, 81};
+
+// Means for the two Gaussians for the six channels (noise)
+const WebRtc_Word16 kNoiseDataMeans[12] = {6738, 4892, 7065, 6715, 6771, 3369, 7646, 3863,
+        7820, 7266, 5020, 4362};
+
+// Means for the two Gaussians for the six channels (speech)
+const WebRtc_Word16 kSpeechDataMeans[12] = {8306, 10085, 10078, 11823, 11843, 6309, 9473,
+        9571, 10879, 7581, 8180, 7483};
+
+// Stds for the two Gaussians for the six channels (noise)
+const WebRtc_Word16 kNoiseDataStds[12] = {378, 1064, 493, 582, 688, 593, 474, 697, 475, 688,
+        421, 455};
+
+// Stds for the two Gaussians for the six channels (speech)
+const WebRtc_Word16 kSpeechDataStds[12] = {555, 505, 567, 524, 585, 1231, 509, 828, 492, 1540,
+        1079, 850};

src/common_audio/vad/main/source/vad_const.h

@@ -0,0 +1,59 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This header file includes the declarations of the internally used constants.
+ */
+
+#ifndef WEBRTC_VAD_CONST_H_
+#define WEBRTC_VAD_CONST_H_
+
+#include "typedefs.h"
+
+// TODO(ajm): give these internal-linkage by moving to the appropriate file
+// where possible, and otherwise tag with WebRtcVad_.
+
+// Spectrum Weighting
+extern const WebRtc_Word16 kSpectrumWeight[];
+extern const WebRtc_Word16 kCompVar;
+// Logarithm constant
+extern const WebRtc_Word16 kLogConst;
+extern const WebRtc_Word16 kLog10Const;
+// Q15
+extern const WebRtc_Word16 kNoiseUpdateConst;
+extern const WebRtc_Word16 kSpeechUpdateConst;
+// Q8
+extern const WebRtc_Word16 kBackEta;
+// Coefficients used by WebRtcVad_HpOutput, Q14
+extern const WebRtc_Word16 kHpZeroCoefs[];
+extern const WebRtc_Word16 kHpPoleCoefs[];
+// Allpass filter coefficients, upper and lower, in Q15 resp. Q13
+extern const WebRtc_Word16 kAllPassCoefsQ15[];
+extern const WebRtc_Word16 kAllPassCoefsQ13[];
+// Minimum difference between the two models, Q5
+extern const WebRtc_Word16 kMinimumDifference[];
+// Maximum value when updating the speech model, Q7
+extern const WebRtc_Word16 kMaximumSpeech[];
+// Minimum value for mean value
+extern const WebRtc_Word16 kMinimumMean[];
+// Upper limit of mean value for noise model, Q7
+extern const WebRtc_Word16 kMaximumNoise[];
+// Adjustment for division with two in WebRtcVad_SplitFilter
+extern const WebRtc_Word16 kOffsetVector[];
+// Start values for the Gaussian models, Q7
+extern const WebRtc_Word16 kNoiseDataWeights[];
+extern const WebRtc_Word16 kSpeechDataWeights[];
+extern const WebRtc_Word16 kNoiseDataMeans[];
+extern const WebRtc_Word16 kSpeechDataMeans[];
+extern const WebRtc_Word16 kNoiseDataStds[];
+extern const WebRtc_Word16 kSpeechDataStds[];
+
+#endif // WEBRTC_VAD_CONST_H_

src/common_audio/vad/main/source/vad_core.c

@@ -0,0 +1,685 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file includes the implementation of the core functionality in VAD.
+ * For function description, see vad_core.h.
+ */
+
+#include "vad_core.h"
+#include "vad_const.h"
+#include "vad_defines.h"
+#include "vad_filterbank.h"
+#include "vad_gmm.h"
+#include "vad_sp.h"
+#include "signal_processing_library.h"
+
+static const int kInitCheck = 42;
+
+// Initialize VAD
+int WebRtcVad_InitCore(VadInstT *inst, short mode)
+{
+    int i;
+
+    // Initialization of struct
+    inst->vad = 1;
+    inst->frame_counter = 0;
+    inst->over_hang = 0;
+    inst->num_of_speech = 0;
+
+    // Initialization of downsampling filter state
+    inst->downsampling_filter_states[0] = 0;
+    inst->downsampling_filter_states[1] = 0;
+    inst->downsampling_filter_states[2] = 0;
+    inst->downsampling_filter_states[3] = 0;
+
+    // Read initial PDF parameters
+    for (i = 0; i < NUM_TABLE_VALUES; i++)
+    {
+        inst->noise_means[i] = kNoiseDataMeans[i];
+        inst->speech_means[i] = kSpeechDataMeans[i];
+        inst->noise_stds[i] = kNoiseDataStds[i];
+        inst->speech_stds[i] = kSpeechDataStds[i];
+    }
+
+    // Index and Minimum value vectors are initialized
+    for (i = 0; i < 16 * NUM_CHANNELS; i++)
+    {
+        inst->low_value_vector[i] = 10000;
+        inst->index_vector[i] = 0;
+    }
+
+    for (i = 0; i < 5; i++)
+    {
+        inst->upper_state[i] = 0;
+        inst->lower_state[i] = 0;
+    }
+
+    for (i = 0; i < 4; i++)
+    {
+        inst->hp_filter_state[i] = 0;
+    }
+
+    // Init mean value memory, for FindMin function
+    inst->mean_value[0] = 1600;
+    inst->mean_value[1] = 1600;
+    inst->mean_value[2] = 1600;
+    inst->mean_value[3] = 1600;
+    inst->mean_value[4] = 1600;
+    inst->mean_value[5] = 1600;
+
+    if (mode == 0)
+    {
+        // Quality mode
+        inst->over_hang_max_1[0] = OHMAX1_10MS_Q; // Overhang short speech burst
+        inst->over_hang_max_1[1] = OHMAX1_20MS_Q; // Overhang short speech burst
+        inst->over_hang_max_1[2] = OHMAX1_30MS_Q; // Overhang short speech burst
+        inst->over_hang_max_2[0] = OHMAX2_10MS_Q; // Overhang long speech burst
+        inst->over_hang_max_2[1] = OHMAX2_20MS_Q; // Overhang long speech burst
+        inst->over_hang_max_2[2] = OHMAX2_30MS_Q; // Overhang long speech burst
+
+        inst->individual[0] = INDIVIDUAL_10MS_Q;
+        inst->individual[1] = INDIVIDUAL_20MS_Q;
+        inst->individual[2] = INDIVIDUAL_30MS_Q;
+
+        inst->total[0] = TOTAL_10MS_Q;
+        inst->total[1] = TOTAL_20MS_Q;
+        inst->total[2] = TOTAL_30MS_Q;
+    } else if (mode == 1)
+    {
+        // Low bitrate mode
+        inst->over_hang_max_1[0] = OHMAX1_10MS_LBR; // Overhang short speech burst
+        inst->over_hang_max_1[1] = OHMAX1_20MS_LBR; // Overhang short speech burst
+        inst->over_hang_max_1[2] = OHMAX1_30MS_LBR; // Overhang short speech burst
+        inst->over_hang_max_2[0] = OHMAX2_10MS_LBR; // Overhang long speech burst
+        inst->over_hang_max_2[1] = OHMAX2_20MS_LBR; // Overhang long speech burst
+        inst->over_hang_max_2[2] = OHMAX2_30MS_LBR; // Overhang long speech burst
+
+        inst->individual[0] = INDIVIDUAL_10MS_LBR;
+        inst->individual[1] = INDIVIDUAL_20MS_LBR;
+        inst->individual[2] = INDIVIDUAL_30MS_LBR;
+
+        inst->total[0] = TOTAL_10MS_LBR;
+        inst->total[1] = TOTAL_20MS_LBR;
+        inst->total[2] = TOTAL_30MS_LBR;
+    } else if (mode == 2)
+    {
+        // Aggressive mode
+        inst->over_hang_max_1[0] = OHMAX1_10MS_AGG; // Overhang short speech burst
+        inst->over_hang_max_1[1] = OHMAX1_20MS_AGG; // Overhang short speech burst
+        inst->over_hang_max_1[2] = OHMAX1_30MS_AGG; // Overhang short speech burst
+        inst->over_hang_max_2[0] = OHMAX2_10MS_AGG; // Overhang long speech burst
+        inst->over_hang_max_2[1] = OHMAX2_20MS_AGG; // Overhang long speech burst
+        inst->over_hang_max_2[2] = OHMAX2_30MS_AGG; // Overhang long speech burst
+
+        inst->individual[0] = INDIVIDUAL_10MS_AGG;
+        inst->individual[1] = INDIVIDUAL_20MS_AGG;
+        inst->individual[2] = INDIVIDUAL_30MS_AGG;
+
+        inst->total[0] = TOTAL_10MS_AGG;
+        inst->total[1] = TOTAL_20MS_AGG;
+        inst->total[2] = TOTAL_30MS_AGG;
+    } else
+    {
+        // Very aggressive mode
+        inst->over_hang_max_1[0] = OHMAX1_10MS_VAG; // Overhang short speech burst
+        inst->over_hang_max_1[1] = OHMAX1_20MS_VAG; // Overhang short speech burst
+        inst->over_hang_max_1[2] = OHMAX1_30MS_VAG; // Overhang short speech burst
+        inst->over_hang_max_2[0] = OHMAX2_10MS_VAG; // Overhang long speech burst
+        inst->over_hang_max_2[1] = OHMAX2_20MS_VAG; // Overhang long speech burst
+        inst->over_hang_max_2[2] = OHMAX2_30MS_VAG; // Overhang long speech burst
+
+        inst->individual[0] = INDIVIDUAL_10MS_VAG;
+        inst->individual[1] = INDIVIDUAL_20MS_VAG;
+        inst->individual[2] = INDIVIDUAL_30MS_VAG;
+
+        inst->total[0] = TOTAL_10MS_VAG;
+        inst->total[1] = TOTAL_20MS_VAG;
+        inst->total[2] = TOTAL_30MS_VAG;
+    }
+
+    inst->init_flag = kInitCheck;
+
+    return 0;
+}
+
+// Set aggressiveness mode
+int WebRtcVad_set_mode_core(VadInstT *inst, short mode)
+{
+
+    if (mode == 0)
+    {
+        // Quality mode
+        inst->over_hang_max_1[0] = OHMAX1_10MS_Q; // Overhang short speech burst
+        inst->over_hang_max_1[1] = OHMAX1_20MS_Q; // Overhang short speech burst
+        inst->over_hang_max_1[2] = OHMAX1_30MS_Q; // Overhang short speech burst
+        inst->over_hang_max_2[0] = OHMAX2_10MS_Q; // Overhang long speech burst
+        inst->over_hang_max_2[1] = OHMAX2_20MS_Q; // Overhang long speech burst
+        inst->over_hang_max_2[2] = OHMAX2_30MS_Q; // Overhang long speech burst
+
+        inst->individual[0] = INDIVIDUAL_10MS_Q;
+        inst->individual[1] = INDIVIDUAL_20MS_Q;
+        inst->individual[2] = INDIVIDUAL_30MS_Q;
+
+        inst->total[0] = TOTAL_10MS_Q;
+        inst->total[1] = TOTAL_20MS_Q;
+        inst->total[2] = TOTAL_30MS_Q;
+    } else if (mode == 1)
+    {
+        // Low bitrate mode
+        inst->over_hang_max_1[0] = OHMAX1_10MS_LBR; // Overhang short speech burst
+        inst->over_hang_max_1[1] = OHMAX1_20MS_LBR; // Overhang short speech burst
+        inst->over_hang_max_1[2] = OHMAX1_30MS_LBR; // Overhang short speech burst
+        inst->over_hang_max_2[0] = OHMAX2_10MS_LBR; // Overhang long speech burst
+        inst->over_hang_max_2[1] = OHMAX2_20MS_LBR; // Overhang long speech burst
+        inst->over_hang_max_2[2] = OHMAX2_30MS_LBR; // Overhang long speech burst
+
+        inst->individual[0] = INDIVIDUAL_10MS_LBR;
+        inst->individual[1] = INDIVIDUAL_20MS_LBR;
+        inst->individual[2] = INDIVIDUAL_30MS_LBR;
+
+        inst->total[0] = TOTAL_10MS_LBR;
+        inst->total[1] = TOTAL_20MS_LBR;
+        inst->total[2] = TOTAL_30MS_LBR;
+    } else if (mode == 2)
+    {
+        // Aggressive mode
+        inst->over_hang_max_1[0] = OHMAX1_10MS_AGG; // Overhang short speech burst
+        inst->over_hang_max_1[1] = OHMAX1_20MS_AGG; // Overhang short speech burst
+        inst->over_hang_max_1[2] = OHMAX1_30MS_AGG; // Overhang short speech burst
+        inst->over_hang_max_2[0] = OHMAX2_10MS_AGG; // Overhang long speech burst
+        inst->over_hang_max_2[1] = OHMAX2_20MS_AGG; // Overhang long speech burst
+        inst->over_hang_max_2[2] = OHMAX2_30MS_AGG; // Overhang long speech burst
+
+        inst->individual[0] = INDIVIDUAL_10MS_AGG;
+        inst->individual[1] = INDIVIDUAL_20MS_AGG;
+        inst->individual[2] = INDIVIDUAL_30MS_AGG;
+
+        inst->total[0] = TOTAL_10MS_AGG;
+        inst->total[1] = TOTAL_20MS_AGG;
+        inst->total[2] = TOTAL_30MS_AGG;
+    } else if (mode == 3)
+    {
+        // Very aggressive mode
+        inst->over_hang_max_1[0] = OHMAX1_10MS_VAG; // Overhang short speech burst
+        inst->over_hang_max_1[1] = OHMAX1_20MS_VAG; // Overhang short speech burst
+        inst->over_hang_max_1[2] = OHMAX1_30MS_VAG; // Overhang short speech burst
+        inst->over_hang_max_2[0] = OHMAX2_10MS_VAG; // Overhang long speech burst
+        inst->over_hang_max_2[1] = OHMAX2_20MS_VAG; // Overhang long speech burst
+        inst->over_hang_max_2[2] = OHMAX2_30MS_VAG; // Overhang long speech burst
+
+        inst->individual[0] = INDIVIDUAL_10MS_VAG;
+        inst->individual[1] = INDIVIDUAL_20MS_VAG;
+        inst->individual[2] = INDIVIDUAL_30MS_VAG;
+
+        inst->total[0] = TOTAL_10MS_VAG;
+        inst->total[1] = TOTAL_20MS_VAG;
+        inst->total[2] = TOTAL_30MS_VAG;
+    } else
+    {
+        return -1;
+    }
+
+    return 0;
+}
+
+// Calculate VAD decision by first extracting feature values and then calculate
+// probability for both speech and background noise.
+
+WebRtc_Word16 WebRtcVad_CalcVad32khz(VadInstT *inst, WebRtc_Word16 *speech_frame,
+                                     int frame_length)
+{
+    WebRtc_Word16 len, vad;
+    WebRtc_Word16 speechWB[480]; // Downsampled speech frame: 960 samples (30ms in SWB)
+    WebRtc_Word16 speechNB[240]; // Downsampled speech frame: 480 samples (30ms in WB)
+
+
+    // Downsample signal 32->16->8 before doing VAD
+    WebRtcVad_Downsampling(speech_frame, speechWB, &(inst->downsampling_filter_states[2]),
+                           frame_length);
+    len = WEBRTC_SPL_RSHIFT_W16(frame_length, 1);
+
+    WebRtcVad_Downsampling(speechWB, speechNB, inst->downsampling_filter_states, len);
+    len = WEBRTC_SPL_RSHIFT_W16(len, 1);
+
+    // Do VAD on an 8 kHz signal
+    vad = WebRtcVad_CalcVad8khz(inst, speechNB, len);
+
+    return vad;
+}
+
+WebRtc_Word16 WebRtcVad_CalcVad16khz(VadInstT *inst, WebRtc_Word16 *speech_frame,
+                                     int frame_length)
+{
+    WebRtc_Word16 len, vad;
+    WebRtc_Word16 speechNB[240]; // Downsampled speech frame: 480 samples (30ms in WB)
+
+    // Wideband: Downsample signal before doing VAD
+    WebRtcVad_Downsampling(speech_frame, speechNB, inst->downsampling_filter_states,
+                           frame_length);
+
+    len = WEBRTC_SPL_RSHIFT_W16(frame_length, 1);
+    vad = WebRtcVad_CalcVad8khz(inst, speechNB, len);
+
+    return vad;
+}
+
+WebRtc_Word16 WebRtcVad_CalcVad8khz(VadInstT *inst, WebRtc_Word16 *speech_frame,
+                                    int frame_length)
+{
+    WebRtc_Word16 feature_vector[NUM_CHANNELS], total_power;
+
+    // Get power in the bands
+    total_power = WebRtcVad_get_features(inst, speech_frame, frame_length, feature_vector);
+
+    // Make a VAD
+    inst->vad = WebRtcVad_GmmProbability(inst, feature_vector, total_power, frame_length);
+
+    return inst->vad;
+}
+
+// Calculate probability for both speech and background noise, and perform a
+// hypothesis-test.
+WebRtc_Word16 WebRtcVad_GmmProbability(VadInstT *inst, WebRtc_Word16 *feature_vector,
+                                       WebRtc_Word16 total_power, int frame_length)
+{
+    int n, k;
+    WebRtc_Word16 backval;
+    WebRtc_Word16 h0, h1;
+    WebRtc_Word16 ratvec, xval;
+    WebRtc_Word16 vadflag;
+    WebRtc_Word16 shifts0, shifts1;
+    WebRtc_Word16 tmp16, tmp16_1, tmp16_2;
+    WebRtc_Word16 diff, nr, pos;
+    WebRtc_Word16 nmk, nmk2, nmk3, smk, smk2, nsk, ssk;
+    WebRtc_Word16 delt, ndelt;
+    WebRtc_Word16 maxspe, maxmu;
+    WebRtc_Word16 deltaN[NUM_TABLE_VALUES], deltaS[NUM_TABLE_VALUES];
+    WebRtc_Word16 ngprvec[NUM_TABLE_VALUES], sgprvec[NUM_TABLE_VALUES];
+    WebRtc_Word32 h0test, h1test;
+    WebRtc_Word32 tmp32_1, tmp32_2;
+    WebRtc_Word32 dotVal;
+    WebRtc_Word32 nmid, smid;
+    WebRtc_Word32 probn[NUM_MODELS], probs[NUM_MODELS];
+    WebRtc_Word16 *nmean1ptr, *nmean2ptr, *smean1ptr, *smean2ptr, *nstd1ptr, *nstd2ptr,
+            *sstd1ptr, *sstd2ptr;
+    WebRtc_Word16 overhead1, overhead2, individualTest, totalTest;
+
+    // Set the thresholds to different values based on frame length
+    if (frame_length == 80)
+    {
+        // 80 input samples
+        overhead1 = inst->over_hang_max_1[0];
+        overhead2 = inst->over_hang_max_2[0];
+        individualTest = inst->individual[0];
+        totalTest = inst->total[0];
+    } else if (frame_length == 160)
+    {
+        // 160 input samples
+        overhead1 = inst->over_hang_max_1[1];
+        overhead2 = inst->over_hang_max_2[1];
+        individualTest = inst->individual[1];
+        totalTest = inst->total[1];
+    } else
+    {
+        // 240 input samples
+        overhead1 = inst->over_hang_max_1[2];
+        overhead2 = inst->over_hang_max_2[2];
+        individualTest = inst->individual[2];
+        totalTest = inst->total[2];
+    }
+
+    if (total_power > MIN_ENERGY)
+    { // If signal present at all
+
+        // Set pointers to the gaussian parameters
+        nmean1ptr = &inst->noise_means[0];
+        nmean2ptr = &inst->noise_means[NUM_CHANNELS];
+        smean1ptr = &inst->speech_means[0];
+        smean2ptr = &inst->speech_means[NUM_CHANNELS];
+        nstd1ptr = &inst->noise_stds[0];
+        nstd2ptr = &inst->noise_stds[NUM_CHANNELS];
+        sstd1ptr = &inst->speech_stds[0];
+        sstd2ptr = &inst->speech_stds[NUM_CHANNELS];
+
+        vadflag = 0;
+        dotVal = 0;
+        for (n = 0; n < NUM_CHANNELS; n++)
+        { // For all channels
+
+            pos = WEBRTC_SPL_LSHIFT_W16(n, 1);
+            xval = feature_vector[n];
+
+            // Probability for Noise, Q7 * Q20 = Q27
+            tmp32_1 = WebRtcVad_GaussianProbability(xval, *nmean1ptr++, *nstd1ptr++,
+                                                    &deltaN[pos]);
+            probn[0] = (WebRtc_Word32)(kNoiseDataWeights[n] * tmp32_1);
+            tmp32_1 = WebRtcVad_GaussianProbability(xval, *nmean2ptr++, *nstd2ptr++,
+                                                    &deltaN[pos + 1]);
+            probn[1] = (WebRtc_Word32)(kNoiseDataWeights[n + NUM_CHANNELS] * tmp32_1);
+            h0test = probn[0] + probn[1]; // Q27
+            h0 = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(h0test, 12); // Q15
+
+            // Probability for Speech
+            tmp32_1 = WebRtcVad_GaussianProbability(xval, *smean1ptr++, *sstd1ptr++,
+                                                    &deltaS[pos]);
+            probs[0] = (WebRtc_Word32)(kSpeechDataWeights[n] * tmp32_1);
+            tmp32_1 = WebRtcVad_GaussianProbability(xval, *smean2ptr++, *sstd2ptr++,
+                                                    &deltaS[pos + 1]);
+            probs[1] = (WebRtc_Word32)(kSpeechDataWeights[n + NUM_CHANNELS] * tmp32_1);
+            h1test = probs[0] + probs[1]; // Q27
+            h1 = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(h1test, 12); // Q15
+
+            // Get likelihood ratio. Approximate log2(H1/H0) with shifts0 - shifts1
+            shifts0 = WebRtcSpl_NormW32(h0test);
+            shifts1 = WebRtcSpl_NormW32(h1test);
+
+            if ((h0test > 0) && (h1test > 0))
+            {
+                ratvec = shifts0 - shifts1;
+            } else if (h1test > 0)
+            {
+                ratvec = 31 - shifts1;
+            } else if (h0test > 0)
+            {
+                ratvec = shifts0 - 31;
+            } else
+            {
+                ratvec = 0;
+            }
+
+            // VAD decision with spectrum weighting
+            dotVal += WEBRTC_SPL_MUL_16_16(ratvec, kSpectrumWeight[n]);
+
+            // Individual channel test
+            if ((ratvec << 2) > individualTest)
+            {
+                vadflag = 1;
+            }
+
+            // Probabilities used when updating model
+            if (h0 > 0)
+            {
+                tmp32_1 = probn[0] & 0xFFFFF000; // Q27
+                tmp32_2 = WEBRTC_SPL_LSHIFT_W32(tmp32_1, 2); // Q29
+                ngprvec[pos] = (WebRtc_Word16)WebRtcSpl_DivW32W16(tmp32_2, h0);
+                ngprvec[pos + 1] = 16384 - ngprvec[pos];
+            } else
+            {
+                ngprvec[pos] = 16384;
+                ngprvec[pos + 1] = 0;
+            }
+
+            // Probabilities used when updating model
+            if (h1 > 0)
+            {
+                tmp32_1 = probs[0] & 0xFFFFF000;
+                tmp32_2 = WEBRTC_SPL_LSHIFT_W32(tmp32_1, 2);
+                sgprvec[pos] = (WebRtc_Word16)WebRtcSpl_DivW32W16(tmp32_2, h1);
+                sgprvec[pos + 1] = 16384 - sgprvec[pos];
+            } else
+            {
+                sgprvec[pos] = 0;
+                sgprvec[pos + 1] = 0;
+            }
+        }
+
+        // Overall test
+        if (dotVal >= totalTest)
+        {
+            vadflag |= 1;
+        }
+
+        // Set pointers to the means and standard deviations.
+        nmean1ptr = &inst->noise_means[0];
+        smean1ptr = &inst->speech_means[0];
+        nstd1ptr = &inst->noise_stds[0];
+        sstd1ptr = &inst->speech_stds[0];
+
+        maxspe = 12800;
+
+        // Update the model's parameters
+        for (n = 0; n < NUM_CHANNELS; n++)
+        {
+
+            pos = WEBRTC_SPL_LSHIFT_W16(n, 1);
+
+            // Get min value in past which is used for long term correction
+            backval = WebRtcVad_FindMinimum(inst, feature_vector[n], n); // Q4
+
+            // Compute the "global" mean, that is the sum of the two means weighted
+            nmid = WEBRTC_SPL_MUL_16_16(kNoiseDataWeights[n], *nmean1ptr); // Q7 * Q7
+            nmid += WEBRTC_SPL_MUL_16_16(kNoiseDataWeights[n+NUM_CHANNELS],
+                    *(nmean1ptr+NUM_CHANNELS));
+            tmp16_1 = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(nmid, 6); // Q8
+
+            for (k = 0; k < NUM_MODELS; k++)
+            {
+
+                nr = pos + k;
+
+                nmean2ptr = nmean1ptr + k * NUM_CHANNELS;
+                smean2ptr = smean1ptr + k * NUM_CHANNELS;
+                nstd2ptr = nstd1ptr + k * NUM_CHANNELS;
+                sstd2ptr = sstd1ptr + k * NUM_CHANNELS;
+                nmk = *nmean2ptr;
+                smk = *smean2ptr;
+                nsk = *nstd2ptr;
+                ssk = *sstd2ptr;
+
+                // Update noise mean vector if the frame consists of noise only
+                nmk2 = nmk;
+                if (!vadflag)
+                {
+                    // deltaN = (x-mu)/sigma^2
+                    // ngprvec[k] = probn[k]/(probn[0] + probn[1])
+
+                    delt = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(ngprvec[nr],
+                            deltaN[nr], 11); // Q14*Q11
+                    nmk2 = nmk + (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(delt,
+                            kNoiseUpdateConst,
+                            22); // Q7+(Q14*Q15>>22)
+                }
+
+                // Long term correction of the noise mean
+                ndelt = WEBRTC_SPL_LSHIFT_W16(backval, 4);
+                ndelt -= tmp16_1; // Q8 - Q8
+                nmk3 = nmk2 + (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(ndelt,
+                        kBackEta,
+                        9); // Q7+(Q8*Q8)>>9
+
+                // Control that the noise mean does not drift to much
+                tmp16 = WEBRTC_SPL_LSHIFT_W16(k+5, 7);
+                if (nmk3 < tmp16)
+                    nmk3 = tmp16;
+                tmp16 = WEBRTC_SPL_LSHIFT_W16(72+k-n, 7);
+                if (nmk3 > tmp16)
+                    nmk3 = tmp16;
+                *nmean2ptr = nmk3;
+
+                if (vadflag)
+                {
+                    // Update speech mean vector:
+                    // deltaS = (x-mu)/sigma^2
+                    // sgprvec[k] = probn[k]/(probn[0] + probn[1])
+
+                    delt = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(sgprvec[nr],
+                            deltaS[nr],
+                            11); // (Q14*Q11)>>11=Q14
+                    tmp16 = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(delt,
+                            kSpeechUpdateConst,
+                            21) + 1;
+                    smk2 = smk + (tmp16 >> 1); // Q7 + (Q14 * Q15 >> 22)
+
+                    // Control that the speech mean does not drift to much
+                    maxmu = maxspe + 640;
+                    if (smk2 < kMinimumMean[k])
+                        smk2 = kMinimumMean[k];
+                    if (smk2 > maxmu)
+                        smk2 = maxmu;
+
+                    *smean2ptr = smk2;
+
+                    // (Q7>>3) = Q4
+                    tmp16 = WEBRTC_SPL_RSHIFT_W16((smk + 4), 3);
+
+                    tmp16 = feature_vector[n] - tmp16; // Q4
+                    tmp32_1 = WEBRTC_SPL_MUL_16_16_RSFT(deltaS[nr], tmp16, 3);
+                    tmp32_2 = tmp32_1 - (WebRtc_Word32)4096; // Q12
+                    tmp16 = WEBRTC_SPL_RSHIFT_W16((sgprvec[nr]), 2);
+                    tmp32_1 = (WebRtc_Word32)(tmp16 * tmp32_2);// (Q15>>3)*(Q14>>2)=Q12*Q12=Q24
+
+                    tmp32_2 = WEBRTC_SPL_RSHIFT_W32(tmp32_1, 4); // Q20
+
+                    // 0.1 * Q20 / Q7 = Q13
+                    if (tmp32_2 > 0)
+                        tmp16 = (WebRtc_Word16)WebRtcSpl_DivW32W16(tmp32_2, ssk * 10);
+                    else
+                    {
+                        tmp16 = (WebRtc_Word16)WebRtcSpl_DivW32W16(-tmp32_2, ssk * 10);
+                        tmp16 = -tmp16;
+                    }
+                    // divide by 4 giving an update factor of 0.025
+                    tmp16 += 128; // Rounding
+                    ssk += WEBRTC_SPL_RSHIFT_W16(tmp16, 8);
+                    // Division with 8 plus Q7
+                    if (ssk < MIN_STD)
+                        ssk = MIN_STD;
+                    *sstd2ptr = ssk;
+                } else
+                {
+                    // Update GMM variance vectors
+                    // deltaN * (feature_vector[n] - nmk) - 1, Q11 * Q4
+                    tmp16 = feature_vector[n] - WEBRTC_SPL_RSHIFT_W16(nmk, 3);
+
+                    // (Q15>>3) * (Q14>>2) = Q12 * Q12 = Q24
+                    tmp32_1 = WEBRTC_SPL_MUL_16_16_RSFT(deltaN[nr], tmp16, 3) - 4096;
+                    tmp16 = WEBRTC_SPL_RSHIFT_W16((ngprvec[nr]+2), 2);
+                    tmp32_2 = (WebRtc_Word32)(tmp16 * tmp32_1);
+                    tmp32_1 = WEBRTC_SPL_RSHIFT_W32(tmp32_2, 14);
+                    // Q20  * approx 0.001 (2^-10=0.0009766)
+
+                    // Q20 / Q7 = Q13
+                    tmp16 = (WebRtc_Word16)WebRtcSpl_DivW32W16(tmp32_1, nsk);
+                    if (tmp32_1 > 0)
+                        tmp16 = (WebRtc_Word16)WebRtcSpl_DivW32W16(tmp32_1, nsk);
+                    else
+                    {
+                        tmp16 = (WebRtc_Word16)WebRtcSpl_DivW32W16(-tmp32_1, nsk);
+                        tmp16 = -tmp16;
+                    }
+                    tmp16 += 32; // Rounding
+                    nsk += WEBRTC_SPL_RSHIFT_W16(tmp16, 6);
+
+                    if (nsk < MIN_STD)
+                        nsk = MIN_STD;
+
+                    *nstd2ptr = nsk;
+                }
+            }
+
+            // Separate models if they are too close - nmid in Q14
+            nmid = WEBRTC_SPL_MUL_16_16(kNoiseDataWeights[n], *nmean1ptr);
+            nmid += WEBRTC_SPL_MUL_16_16(kNoiseDataWeights[n+NUM_CHANNELS], *nmean2ptr);
+
+            // smid in Q14
+            smid = WEBRTC_SPL_MUL_16_16(kSpeechDataWeights[n], *smean1ptr);
+            smid += WEBRTC_SPL_MUL_16_16(kSpeechDataWeights[n+NUM_CHANNELS], *smean2ptr);
+
+            // diff = "global" speech mean - "global" noise mean
+            diff = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(smid, 9);
+            tmp16 = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(nmid, 9);
+            diff -= tmp16;
+
+            if (diff < kMinimumDifference[n])
+            {
+
+                tmp16 = kMinimumDifference[n] - diff; // Q5
+
+                // tmp16_1 = ~0.8 * (kMinimumDifference - diff) in Q7
+                // tmp16_2 = ~0.2 * (kMinimumDifference - diff) in Q7
+                tmp16_1 = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(13, tmp16, 2);
+                tmp16_2 = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(3, tmp16, 2);
+
+                // First Gauss, speech model
+                tmp16 = tmp16_1 + *smean1ptr;
+                *smean1ptr = tmp16;
+                smid = WEBRTC_SPL_MUL_16_16(tmp16, kSpeechDataWeights[n]);
+
+                // Second Gauss, speech model
+                tmp16 = tmp16_1 + *smean2ptr;
+                *smean2ptr = tmp16;
+                smid += WEBRTC_SPL_MUL_16_16(tmp16, kSpeechDataWeights[n+NUM_CHANNELS]);
+
+                // First Gauss, noise model
+                tmp16 = *nmean1ptr - tmp16_2;
+                *nmean1ptr = tmp16;
+
+                nmid = WEBRTC_SPL_MUL_16_16(tmp16, kNoiseDataWeights[n]);
+
+                // Second Gauss, noise model
+                tmp16 = *nmean2ptr - tmp16_2;
+                *nmean2ptr = tmp16;
+                nmid += WEBRTC_SPL_MUL_16_16(tmp16, kNoiseDataWeights[n+NUM_CHANNELS]);
+            }
+
+            // Control that the speech & noise means do not drift to much
+            maxspe = kMaximumSpeech[n];
+            tmp16_2 = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(smid, 7);
+            if (tmp16_2 > maxspe)
+            { // Upper limit of speech model
+                tmp16_2 -= maxspe;
+
+                *smean1ptr -= tmp16_2;
+                *smean2ptr -= tmp16_2;
+            }
+
+            tmp16_2 = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(nmid, 7);
+            if (tmp16_2 > kMaximumNoise[n])
+            {
+                tmp16_2 -= kMaximumNoise[n];
+
+                *nmean1ptr -= tmp16_2;
+                *nmean2ptr -= tmp16_2;
+            }
+
+            nmean1ptr++;
+            smean1ptr++;
+            nstd1ptr++;
+            sstd1ptr++;
+        }
+        inst->frame_counter++;
+    } else
+    {
+        vadflag = 0;
+    }
+
+    // Hangover smoothing
+    if (!vadflag)
+    {
+        if (inst->over_hang > 0)
+        {
+            vadflag = 2 + inst->over_hang;
+            inst->over_hang = inst->over_hang - 1;
+        }
+        inst->num_of_speech = 0;
+    } else
+    {
+        inst->num_of_speech = inst->num_of_speech + 1;
+        if (inst->num_of_speech > NSP_MAX)
+        {
+            inst->num_of_speech = NSP_MAX;
+            inst->over_hang = overhead2;
+        } else
+            inst->over_hang = overhead1;
+    }
+    return vadflag;
+}

src/common_audio/vad/main/source/vad_core.h

@@ -0,0 +1,132 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This header file includes the descriptions of the core VAD calls.
+ */
+
+#ifndef WEBRTC_VAD_CORE_H_
+#define WEBRTC_VAD_CORE_H_
+
+#include "typedefs.h"
+#include "vad_defines.h"
+
+typedef struct VadInstT_
+{
+
+    WebRtc_Word16 vad;
+    WebRtc_Word32 downsampling_filter_states[4];
+    WebRtc_Word16 noise_means[NUM_TABLE_VALUES];
+    WebRtc_Word16 speech_means[NUM_TABLE_VALUES];
+    WebRtc_Word16 noise_stds[NUM_TABLE_VALUES];
+    WebRtc_Word16 speech_stds[NUM_TABLE_VALUES];
+    WebRtc_Word32 frame_counter;
+    WebRtc_Word16 over_hang; // Over Hang
+    WebRtc_Word16 num_of_speech;
+    WebRtc_Word16 index_vector[16 * NUM_CHANNELS];
+    WebRtc_Word16 low_value_vector[16 * NUM_CHANNELS];
+    WebRtc_Word16 mean_value[NUM_CHANNELS];
+    WebRtc_Word16 upper_state[5];
+    WebRtc_Word16 lower_state[5];
+    WebRtc_Word16 hp_filter_state[4];
+    WebRtc_Word16 over_hang_max_1[3];
+    WebRtc_Word16 over_hang_max_2[3];
+    WebRtc_Word16 individual[3];
+    WebRtc_Word16 total[3];
+
+    short init_flag;
+
+} VadInstT;
+
+/****************************************************************************
+ * WebRtcVad_InitCore(...)
+ *
+ * This function initializes a VAD instance
+ *
+ * Input:
+ *      - inst      : Instance that should be initialized
+ *      - mode      : Aggressiveness degree
+ *                    0 (High quality) - 3 (Highly aggressive)
+ *
+ * Output:
+ *      - inst      : Initialized instance
+ *
+ * Return value     :  0 - Ok
+ *                    -1 - Error
+ */
+int WebRtcVad_InitCore(VadInstT* inst, short mode);
+
+/****************************************************************************
+ * WebRtcVad_set_mode_core(...)
+ *
+ * This function changes the VAD settings
+ *
+ * Input:
+ *      - inst      : VAD instance
+ *      - mode      : Aggressiveness degree
+ *                    0 (High quality) - 3 (Highly aggressive)
+ *
+ * Output:
+ *      - inst      : Changed  instance
+ *
+ * Return value     :  0 - Ok
+ *                    -1 - Error
+ */
+
+int WebRtcVad_set_mode_core(VadInstT* inst, short mode);
+
+/****************************************************************************
+ * WebRtcVad_CalcVad32khz(...) 
+ * WebRtcVad_CalcVad16khz(...) 
+ * WebRtcVad_CalcVad8khz(...) 
+ *
+ * Calculate probability for active speech and make VAD decision.
+ *
+ * Input:
+ *      - inst          : Instance that should be initialized
+ *      - speech_frame  : Input speech frame
+ *      - frame_length  : Number of input samples
+ *
+ * Output:
+ *      - inst          : Updated filter states etc.
+ *
+ * Return value         : VAD decision
+ *                        0 - No active speech
+ *                        1-6 - Active speech
+ */
+WebRtc_Word16 WebRtcVad_CalcVad32khz(VadInstT* inst, WebRtc_Word16* speech_frame,
+                                     int frame_length);
+WebRtc_Word16 WebRtcVad_CalcVad16khz(VadInstT* inst, WebRtc_Word16* speech_frame,
+                                     int frame_length);
+WebRtc_Word16 WebRtcVad_CalcVad8khz(VadInstT* inst, WebRtc_Word16* speech_frame,
+                                    int frame_length);
+
+/****************************************************************************
+ * WebRtcVad_GmmProbability(...)
+ *
+ * This function calculates the probabilities for background noise and
+ * speech using Gaussian Mixture Models. A hypothesis-test is performed to decide
+ * which type of signal is most probable.
+ *
+ * Input:
+ *      - inst              : Pointer to VAD instance
+ *      - feature_vector    : Feature vector = log10(energy in frequency band)
+ *      - total_power       : Total power in frame.
+ *      - frame_length      : Number of input samples
+ *
+ * Output:
+ *      VAD decision        : 0 - noise, 1 - speech
+ *    
+ */
+WebRtc_Word16 WebRtcVad_GmmProbability(VadInstT* inst, WebRtc_Word16* feature_vector,
+                                       WebRtc_Word16 total_power, int frame_length);
+
+#endif // WEBRTC_VAD_CORE_H_

src/common_audio/vad/main/source/vad_defines.h

@@ -0,0 +1,95 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This header file includes the macros used in VAD.
+ */
+
+#ifndef WEBRTC_VAD_DEFINES_H_
+#define WEBRTC_VAD_DEFINES_H_
+
+#define NUM_CHANNELS        6   // Eight frequency bands
+#define NUM_MODELS          2   // Number of Gaussian models
+#define NUM_TABLE_VALUES    NUM_CHANNELS * NUM_MODELS
+
+#define MIN_ENERGY          10
+#define ALPHA1              6553    // 0.2 in Q15
+#define ALPHA2              32439   // 0.99 in Q15
+#define NSP_MAX             6       // Maximum number of VAD=1 frames in a row counted
+#define MIN_STD             384     // Minimum standard deviation
+// Mode 0, Quality thresholds - Different thresholds for the different frame lengths
+#define INDIVIDUAL_10MS_Q   24
+#define INDIVIDUAL_20MS_Q   21      // (log10(2)*66)<<2 ~=16
+#define INDIVIDUAL_30MS_Q   24
+
+#define TOTAL_10MS_Q        57
+#define TOTAL_20MS_Q        48
+#define TOTAL_30MS_Q        57
+
+#define OHMAX1_10MS_Q       8  // Max Overhang 1
+#define OHMAX2_10MS_Q       14 // Max Overhang 2
+#define OHMAX1_20MS_Q       4  // Max Overhang 1
+#define OHMAX2_20MS_Q       7  // Max Overhang 2
+#define OHMAX1_30MS_Q       3
+#define OHMAX2_30MS_Q       5
+
+// Mode 1, Low bitrate thresholds - Different thresholds for the different frame lengths
+#define INDIVIDUAL_10MS_LBR 37
+#define INDIVIDUAL_20MS_LBR 32
+#define INDIVIDUAL_30MS_LBR 37
+
+#define TOTAL_10MS_LBR      100
+#define TOTAL_20MS_LBR      80
+#define TOTAL_30MS_LBR      100
+
+#define OHMAX1_10MS_LBR     8  // Max Overhang 1
+#define OHMAX2_10MS_LBR     14 // Max Overhang 2
+#define OHMAX1_20MS_LBR     4
+#define OHMAX2_20MS_LBR     7
+
+#define OHMAX1_30MS_LBR     3
+#define OHMAX2_30MS_LBR     5
+
+// Mode 2, Very aggressive thresholds - Different thresholds for the different frame lengths
+#define INDIVIDUAL_10MS_AGG 82
+#define INDIVIDUAL_20MS_AGG 78
+#define INDIVIDUAL_30MS_AGG 82
+
+#define TOTAL_10MS_AGG      285 //580
+#define TOTAL_20MS_AGG      260
+#define TOTAL_30MS_AGG      285
+
+#define OHMAX1_10MS_AGG     6  // Max Overhang 1
+#define OHMAX2_10MS_AGG     9  // Max Overhang 2
+#define OHMAX1_20MS_AGG     3
+#define OHMAX2_20MS_AGG     5
+
+#define OHMAX1_30MS_AGG     2
+#define OHMAX2_30MS_AGG     3
+
+// Mode 3, Super aggressive thresholds - Different thresholds for the different frame lengths
+#define INDIVIDUAL_10MS_VAG 94
+#define INDIVIDUAL_20MS_VAG 94
+#define INDIVIDUAL_30MS_VAG 94
+
+#define TOTAL_10MS_VAG      1100 //1700
+#define TOTAL_20MS_VAG      1050
+#define TOTAL_30MS_VAG      1100
+
+#define OHMAX1_10MS_VAG     6  // Max Overhang 1
+#define OHMAX2_10MS_VAG     9  // Max Overhang 2
+#define OHMAX1_20MS_VAG     3
+#define OHMAX2_20MS_VAG     5
+
+#define OHMAX1_30MS_VAG     2
+#define OHMAX2_30MS_VAG     3
+
+#endif // WEBRTC_VAD_DEFINES_H_

src/common_audio/vad/main/source/vad_filterbank.c

@@ -0,0 +1,267 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file includes the implementation of the internal filterbank associated functions.
+ * For function description, see vad_filterbank.h.
+ */
+
+#include "vad_filterbank.h"
+#include "vad_defines.h"
+#include "vad_const.h"
+#include "signal_processing_library.h"
+
+void WebRtcVad_HpOutput(WebRtc_Word16 *in_vector,
+                        WebRtc_Word16 in_vector_length,
+                        WebRtc_Word16 *out_vector,
+                        WebRtc_Word16 *filter_state)
+{
+    WebRtc_Word16 i, *pi, *outPtr;
+    WebRtc_Word32 tmpW32;
+
+    pi = &in_vector[0];
+    outPtr = &out_vector[0];
+
+    // The sum of the absolute values of the impulse response:
+    // The zero/pole-filter has a max amplification of a single sample of: 1.4546
+    // Impulse response: 0.4047 -0.6179 -0.0266  0.1993  0.1035  -0.0194
+    // The all-zero section has a max amplification of a single sample of: 1.6189
+    // Impulse response: 0.4047 -0.8094  0.4047  0       0        0
+    // The all-pole section has a max amplification of a single sample of: 1.9931
+    // Impulse response: 1.0000  0.4734 -0.1189 -0.2187 -0.0627   0.04532
+
+    for (i = 0; i < in_vector_length; i++)
+    {
+        // all-zero section (filter coefficients in Q14)
+        tmpW32 = (WebRtc_Word32)WEBRTC_SPL_MUL_16_16(kHpZeroCoefs[0], (*pi));
+        tmpW32 += (WebRtc_Word32)WEBRTC_SPL_MUL_16_16(kHpZeroCoefs[1], filter_state[0]);
+        tmpW32 += (WebRtc_Word32)WEBRTC_SPL_MUL_16_16(kHpZeroCoefs[2], filter_state[1]); // Q14
+        filter_state[1] = filter_state[0];
+        filter_state[0] = *pi++;
+
+        // all-pole section
+        tmpW32 -= (WebRtc_Word32)WEBRTC_SPL_MUL_16_16(kHpPoleCoefs[1], filter_state[2]); // Q14
+        tmpW32 -= (WebRtc_Word32)WEBRTC_SPL_MUL_16_16(kHpPoleCoefs[2], filter_state[3]);
+        filter_state[3] = filter_state[2];
+        filter_state[2] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32 (tmpW32, 14);
+        *outPtr++ = filter_state[2];
+    }
+}
+
+void WebRtcVad_Allpass(WebRtc_Word16 *in_vector,
+                       WebRtc_Word16 *out_vector,
+                       WebRtc_Word16 filter_coefficients,
+                       int vector_length,
+                       WebRtc_Word16 *filter_state)
+{
+    // The filter can only cause overflow (in the w16 output variable)
+    // if more than 4 consecutive input numbers are of maximum value and
+    // has the the same sign as the impulse responses first taps.
+    // First 6 taps of the impulse response: 0.6399 0.5905 -0.3779
+    // 0.2418 -0.1547 0.0990
+
+    int n;
+    WebRtc_Word16 tmp16;
+    WebRtc_Word32 tmp32, in32, state32;
+
+    state32 = WEBRTC_SPL_LSHIFT_W32(((WebRtc_Word32)(*filter_state)), 16); // Q31
+
+    for (n = 0; n < vector_length; n++)
+    {
+
+        tmp32 = state32 + WEBRTC_SPL_MUL_16_16(filter_coefficients, (*in_vector));
+        tmp16 = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp32, 16);
+        *out_vector++ = tmp16;
+        in32 = WEBRTC_SPL_LSHIFT_W32(((WebRtc_Word32)(*in_vector)), 14);
+        state32 = in32 - WEBRTC_SPL_MUL_16_16(filter_coefficients, tmp16);
+        state32 = WEBRTC_SPL_LSHIFT_W32(state32, 1);
+        in_vector += 2;
+    }
+
+    *filter_state = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(state32, 16);
+}
+
+void WebRtcVad_SplitFilter(WebRtc_Word16 *in_vector,
+                           WebRtc_Word16 *out_vector_hp,
+                           WebRtc_Word16 *out_vector_lp,
+                           WebRtc_Word16 *upper_state,
+                           WebRtc_Word16 *lower_state,
+                           int in_vector_length)
+{
+    WebRtc_Word16 tmpOut;
+    int k, halflen;
+
+    // Downsampling by 2 and get two branches
+    halflen = WEBRTC_SPL_RSHIFT_W16(in_vector_length, 1);
+
+    // All-pass filtering upper branch
+    WebRtcVad_Allpass(&in_vector[0], out_vector_hp, kAllPassCoefsQ15[0], halflen, upper_state);
+
+    // All-pass filtering lower branch
+    WebRtcVad_Allpass(&in_vector[1], out_vector_lp, kAllPassCoefsQ15[1], halflen, lower_state);
+
+    // Make LP and HP signals
+    for (k = 0; k < halflen; k++)
+    {
+        tmpOut = *out_vector_hp;
+        *out_vector_hp++ -= *out_vector_lp;
+        *out_vector_lp++ += tmpOut;
+    }
+}
+
+WebRtc_Word16 WebRtcVad_get_features(VadInstT *inst,
+                                     WebRtc_Word16 *in_vector,
+                                     int frame_size,
+                                     WebRtc_Word16 *out_vector)
+{
+    int curlen, filtno;
+    WebRtc_Word16 vecHP1[120], vecLP1[120];
+    WebRtc_Word16 vecHP2[60], vecLP2[60];
+    WebRtc_Word16 *ptin;
+    WebRtc_Word16 *hptout, *lptout;
+    WebRtc_Word16 power = 0;
+
+    // Split at 2000 Hz and downsample
+    filtno = 0;
+    ptin = in_vector;
+    hptout = vecHP1;
+    lptout = vecLP1;
+    curlen = frame_size;
+    WebRtcVad_SplitFilter(ptin, hptout, lptout, &inst->upper_state[filtno],
+                  &inst->lower_state[filtno], curlen);
+
+    // Split at 3000 Hz and downsample
+    filtno = 1;
+    ptin = vecHP1;
+    hptout = vecHP2;
+    lptout = vecLP2;
+    curlen = WEBRTC_SPL_RSHIFT_W16(frame_size, 1);
+
+    WebRtcVad_SplitFilter(ptin, hptout, lptout, &inst->upper_state[filtno],
+                  &inst->lower_state[filtno], curlen);
+
+    // Energy in 3000 Hz - 4000 Hz
+    curlen = WEBRTC_SPL_RSHIFT_W16(curlen, 1);
+    WebRtcVad_LogOfEnergy(vecHP2, &out_vector[5], &power, kOffsetVector[5], curlen);
+
+    // Energy in 2000 Hz - 3000 Hz
+    WebRtcVad_LogOfEnergy(vecLP2, &out_vector[4], &power, kOffsetVector[4], curlen);
+
+    // Split at 1000 Hz and downsample
+    filtno = 2;
+    ptin = vecLP1;
+    hptout = vecHP2;
+    lptout = vecLP2;
+    curlen = WEBRTC_SPL_RSHIFT_W16(frame_size, 1);
+    WebRtcVad_SplitFilter(ptin, hptout, lptout, &inst->upper_state[filtno],
+                  &inst->lower_state[filtno], curlen);
+
+    // Energy in 1000 Hz - 2000 Hz
+    curlen = WEBRTC_SPL_RSHIFT_W16(curlen, 1);
+    WebRtcVad_LogOfEnergy(vecHP2, &out_vector[3], &power, kOffsetVector[3], curlen);
+
+    // Split at 500 Hz
+    filtno = 3;
+    ptin = vecLP2;
+    hptout = vecHP1;
+    lptout = vecLP1;
+
+    WebRtcVad_SplitFilter(ptin, hptout, lptout, &inst->upper_state[filtno],
+                  &inst->lower_state[filtno], curlen);
+
+    // Energy in 500 Hz - 1000 Hz
+    curlen = WEBRTC_SPL_RSHIFT_W16(curlen, 1);
+    WebRtcVad_LogOfEnergy(vecHP1, &out_vector[2], &power, kOffsetVector[2], curlen);
+    // Split at 250 Hz
+    filtno = 4;
+    ptin = vecLP1;
+    hptout = vecHP2;
+    lptout = vecLP2;
+
+    WebRtcVad_SplitFilter(ptin, hptout, lptout, &inst->upper_state[filtno],
+                  &inst->lower_state[filtno], curlen);
+
+    // Energy in 250 Hz - 500 Hz
+    curlen = WEBRTC_SPL_RSHIFT_W16(curlen, 1);
+    WebRtcVad_LogOfEnergy(vecHP2, &out_vector[1], &power, kOffsetVector[1], curlen);
+
+    // Remove DC and LFs
+    WebRtcVad_HpOutput(vecLP2, curlen, vecHP1, inst->hp_filter_state);
+
+    // Power in 80 Hz - 250 Hz
+    WebRtcVad_LogOfEnergy(vecHP1, &out_vector[0], &power, kOffsetVector[0], curlen);
+
+    return power;
+}
+
+void WebRtcVad_LogOfEnergy(WebRtc_Word16 *vector,
+                           WebRtc_Word16 *enerlogval,
+                           WebRtc_Word16 *power,
+                           WebRtc_Word16 offset,
+                           int vector_length)
+{
+    WebRtc_Word16 enerSum = 0;
+    WebRtc_Word16 zeros, frac, log2;
+    WebRtc_Word32 energy;
+
+    int shfts = 0, shfts2;
+
+    energy = WebRtcSpl_Energy(vector, vector_length, &shfts);
+
+    if (energy > 0)
+    {
+
+        shfts2 = 16 - WebRtcSpl_NormW32(energy);
+        shfts += shfts2;
+        // "shfts" is the total number of right shifts that has been done to enerSum.
+        enerSum = (WebRtc_Word16)WEBRTC_SPL_SHIFT_W32(energy, -shfts2);
+
+        // Find:
+        // 160*log10(enerSum*2^shfts) = 160*log10(2)*log2(enerSum*2^shfts) =
+        // 160*log10(2)*(log2(enerSum) + log2(2^shfts)) =
+        // 160*log10(2)*(log2(enerSum) + shfts)
+
+        zeros = WebRtcSpl_NormU32(enerSum);
+        frac = (WebRtc_Word16)(((WebRtc_UWord32)((WebRtc_Word32)(enerSum) << zeros)
+                & 0x7FFFFFFF) >> 21);
+        log2 = (WebRtc_Word16)(((31 - zeros) << 10) + frac);
+
+        *enerlogval = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(kLogConst, log2, 19)
+                + (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(shfts, kLogConst, 9);
+
+        if (*enerlogval < 0)
+        {
+            *enerlogval = 0;
+        }
+    } else
+    {
+        *enerlogval = 0;
+        shfts = -15;
+        enerSum = 0;
+    }
+
+    *enerlogval += offset;
+
+    // Total power in frame
+    if (*power <= MIN_ENERGY)
+    {
+        if (shfts > 0)
+        {
+            *power += MIN_ENERGY + 1;
+        } else if (WEBRTC_SPL_SHIFT_W16(enerSum, shfts) > MIN_ENERGY)
+        {
+            *power += MIN_ENERGY + 1;
+        } else
+        {
+            *power += WEBRTC_SPL_SHIFT_W16(enerSum, shfts);
+        }
+    }
+}

src/common_audio/vad/main/source/vad_filterbank.h

@@ -0,0 +1,143 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This header file includes the description of the internal VAD call
+ * WebRtcVad_GaussianProbability.
+ */
+
+#ifndef WEBRTC_VAD_FILTERBANK_H_
+#define WEBRTC_VAD_FILTERBANK_H_
+
+#include "vad_core.h"
+
+/****************************************************************************
+ * WebRtcVad_HpOutput(...)
+ *
+ * This function removes DC from the lowest frequency band
+ *
+ * Input:
+ *      - in_vector         : Samples in the frequency interval 0 - 250 Hz
+ *      - in_vector_length  : Length of input and output vector
+ *      - filter_state      : Current state of the filter
+ *
+ * Output:
+ *      - out_vector        : Samples in the frequency interval 80 - 250 Hz
+ *      - filter_state      : Updated state of the filter
+ *
+ */
+void WebRtcVad_HpOutput(WebRtc_Word16* in_vector,
+                        WebRtc_Word16  in_vector_length,
+                        WebRtc_Word16* out_vector,
+                        WebRtc_Word16* filter_state);
+
+/****************************************************************************
+ * WebRtcVad_Allpass(...)
+ *
+ * This function is used when before splitting a speech file into 
+ * different frequency bands
+ *
+ * Note! Do NOT let the arrays in_vector and out_vector correspond to the same address.
+ *
+ * Input:
+ *      - in_vector             : (Q0)
+ *      - filter_coefficients   : (Q15)
+ *      - vector_length         : Length of input and output vector
+ *      - filter_state          : Current state of the filter (Q(-1))
+ *
+ * Output:
+ *      - out_vector            : Output speech signal (Q(-1))
+ *      - filter_state          : Updated state of the filter (Q(-1))
+ *
+ */
+void WebRtcVad_Allpass(WebRtc_Word16* in_vector,
+                       WebRtc_Word16* outw16,
+                       WebRtc_Word16 filter_coefficients,
+                       int vector_length,
+                       WebRtc_Word16* filter_state);
+
+/****************************************************************************
+ * WebRtcVad_SplitFilter(...)
+ *
+ * This function is used when before splitting a speech file into 
+ * different frequency bands
+ *
+ * Input:
+ *      - in_vector         : Input signal to be split into two frequency bands.
+ *      - upper_state       : Current state of the upper filter
+ *      - lower_state       : Current state of the lower filter
+ *      - in_vector_length  : Length of input vector
+ *
+ * Output:
+ *      - out_vector_hp     : Upper half of the spectrum
+ *      - out_vector_lp     : Lower half of the spectrum
+ *      - upper_state       : Updated state of the upper filter
+ *      - lower_state       : Updated state of the lower filter
+ *
+ */
+void WebRtcVad_SplitFilter(WebRtc_Word16* in_vector,
+                           WebRtc_Word16* out_vector_hp,
+                           WebRtc_Word16* out_vector_lp,
+                           WebRtc_Word16* upper_state,
+                           WebRtc_Word16* lower_state,
+                           int in_vector_length);
+
+/****************************************************************************
+ * WebRtcVad_get_features(...)
+ *
+ * This function is used to get the logarithm of the power of each of the 
+ * 6 frequency bands used by the VAD:
+ *        80 Hz - 250 Hz
+ *        250 Hz - 500 Hz
+ *        500 Hz - 1000 Hz
+ *        1000 Hz - 2000 Hz
+ *        2000 Hz - 3000 Hz
+ *        3000 Hz - 4000 Hz 
+ *
+ * Input:
+ *      - inst        : Pointer to VAD instance
+ *      - in_vector   : Input speech signal
+ *      - frame_size  : Frame size, in number of samples
+ *
+ * Output:
+ *      - out_vector  : 10*log10(power in each freq. band), Q4
+ *    
+ * Return: total power in the signal (NOTE! This value is not exact since it
+ *         is only used in a comparison.
+ */
+WebRtc_Word16 WebRtcVad_get_features(VadInstT* inst,
+                                     WebRtc_Word16* in_vector,
+                                     int frame_size,
+                                     WebRtc_Word16* out_vector);
+
+/****************************************************************************
+ * WebRtcVad_LogOfEnergy(...)
+ *
+ * This function is used to get the logarithm of the power of one frequency band.
+ *
+ * Input:
+ *      - vector            : Input speech samples for one frequency band
+ *      - offset            : Offset value for the current frequency band
+ *      - vector_length     : Length of input vector
+ *
+ * Output:
+ *      - enerlogval        : 10*log10(energy);
+ *      - power             : Update total power in speech frame. NOTE! This value
+ *                            is not exact since it is only used in a comparison.
+ *     
+ */
+void WebRtcVad_LogOfEnergy(WebRtc_Word16* vector,
+                           WebRtc_Word16* enerlogval,
+                           WebRtc_Word16* power,
+                           WebRtc_Word16 offset,
+                           int vector_length);
+
+#endif // WEBRTC_VAD_FILTERBANK_H_

src/common_audio/vad/main/source/vad_gmm.c

@@ -0,0 +1,70 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file includes the implementation of the internal VAD call
+ * WebRtcVad_GaussianProbability. For function description, see vad_gmm.h.
+ */
+
+#include "vad_gmm.h"
+#include "signal_processing_library.h"
+#include "vad_const.h"
+
+WebRtc_Word32 WebRtcVad_GaussianProbability(WebRtc_Word16 in_sample,
+                                            WebRtc_Word16 mean,
+                                            WebRtc_Word16 std,
+                                            WebRtc_Word16 *delta)
+{
+    WebRtc_Word16 tmp16, tmpDiv, tmpDiv2, expVal, tmp16_1, tmp16_2;
+    WebRtc_Word32 tmp32, y32;
+
+    // Calculate tmpDiv=1/std, in Q10
+    tmp32 = (WebRtc_Word32)WEBRTC_SPL_RSHIFT_W16(std,1) + (WebRtc_Word32)131072; // 1 in Q17
+    tmpDiv = (WebRtc_Word16)WebRtcSpl_DivW32W16(tmp32, std); // Q17/Q7 = Q10
+
+    // Calculate tmpDiv2=1/std^2, in Q14
+    tmp16 = WEBRTC_SPL_RSHIFT_W16(tmpDiv, 2); // From Q10 to Q8
+    tmpDiv2 = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(tmp16, tmp16, 2); // (Q8 * Q8)>>2 = Q14
+
+    tmp16 = WEBRTC_SPL_LSHIFT_W16(in_sample, 3); // Q7
+    tmp16 = tmp16 - mean; // Q7 - Q7 = Q7
+
+    // To be used later, when updating noise/speech model
+    // delta = (x-m)/std^2, in Q11
+    *delta = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(tmpDiv2, tmp16, 10); //(Q14*Q7)>>10 = Q11
+
+    // Calculate tmp32=(x-m)^2/(2*std^2), in Q10
+    tmp32 = (WebRtc_Word32)WEBRTC_SPL_MUL_16_16_RSFT(*delta, tmp16, 9); // One shift for /2
+
+    // Calculate expVal ~= exp(-(x-m)^2/(2*std^2)) ~= exp2(-log2(exp(1))*tmp32)
+    if (tmp32 < kCompVar)
+    {
+        // Calculate tmp16 = log2(exp(1))*tmp32 , in Q10
+        tmp16 = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT((WebRtc_Word16)tmp32,
+                                                         kLog10Const, 12);
+        tmp16 = -tmp16;
+        tmp16_2 = (WebRtc_Word16)(0x0400 | (tmp16 & 0x03FF));
+        tmp16_1 = (WebRtc_Word16)(tmp16 ^ 0xFFFF);
+        tmp16 = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W16(tmp16_1, 10);
+        tmp16 += 1;
+        // Calculate expVal=log2(-tmp32), in Q10
+        expVal = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32((WebRtc_Word32)tmp16_2, tmp16);
+
+    } else
+    {
+        expVal = 0;
+    }
+
+    // Calculate y32=(1/std)*exp(-(x-m)^2/(2*std^2)), in Q20
+    y32 = WEBRTC_SPL_MUL_16_16(tmpDiv, expVal); // Q10 * Q10 = Q20
+
+    return y32; // Q20
+}

src/common_audio/vad/main/source/vad_gmm.h

@@ -0,0 +1,47 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This header file includes the description of the internal VAD call
+ * WebRtcVad_GaussianProbability.
+ */
+
+#ifndef WEBRTC_VAD_GMM_H_
+#define WEBRTC_VAD_GMM_H_
+
+#include "typedefs.h"
+
+/****************************************************************************
+ * WebRtcVad_GaussianProbability(...)
+ *
+ * This function calculates the probability for the value 'in_sample', given that in_sample
+ * comes from a normal distribution with mean 'mean' and standard deviation 'std'.
+ *
+ * Input:
+ *      - in_sample     : Input sample in Q4
+ *      - mean          : mean value in the statistical model, Q7
+ *      - std           : standard deviation, Q7
+ *
+ * Output:
+ *
+ *      - delta         : Value used when updating the model, Q11
+ *
+ * Return:
+ *      - out           : out = 1/std * exp(-(x-m)^2/(2*std^2));
+ *                        Probability for x.
+ *
+ */
+WebRtc_Word32 WebRtcVad_GaussianProbability(WebRtc_Word16 in_sample,
+                                            WebRtc_Word16 mean,
+                                            WebRtc_Word16 std,
+                                            WebRtc_Word16 *delta);
+
+#endif // WEBRTC_VAD_GMM_H_

src/common_audio/vad/main/source/vad_sp.c

@@ -0,0 +1,231 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file includes the implementation of the VAD internal calls for Downsampling and
+ * FindMinimum.
+ * For function call descriptions; See vad_sp.h.
+ */
+
+#include "vad_sp.h"
+#include "vad_defines.h"
+#include "vad_const.h"
+#include "signal_processing_library.h"
+
+// Downsampling filter based on the splitting filter and the allpass functions
+// in vad_filterbank.c
+void WebRtcVad_Downsampling(WebRtc_Word16* signal_in,
+                            WebRtc_Word16* signal_out,
+                            WebRtc_Word32* filter_state,
+                            int inlen)
+{
+    WebRtc_Word16 tmp16_1, tmp16_2;
+    WebRtc_Word32 tmp32_1, tmp32_2;
+    int n, halflen;
+
+    // Downsampling by 2 and get two branches
+    halflen = WEBRTC_SPL_RSHIFT_W16(inlen, 1);
+
+    tmp32_1 = filter_state[0];
+    tmp32_2 = filter_state[1];
+
+    // Filter coefficients in Q13, filter state in Q0
+    for (n = 0; n < halflen; n++)
+    {
+        // All-pass filtering upper branch
+        tmp16_1 = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp32_1, 1)
+                + (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT((kAllPassCoefsQ13[0]),
+                                                           *signal_in, 14);
+        *signal_out = tmp16_1;
+        tmp32_1 = (WebRtc_Word32)(*signal_in++)
+                - (WebRtc_Word32)WEBRTC_SPL_MUL_16_16_RSFT((kAllPassCoefsQ13[0]), tmp16_1, 12);
+
+        // All-pass filtering lower branch
+        tmp16_2 = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp32_2, 1)
+                + (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT((kAllPassCoefsQ13[1]),
+                                                           *signal_in, 14);
+        *signal_out++ += tmp16_2;
+        tmp32_2 = (WebRtc_Word32)(*signal_in++)
+                - (WebRtc_Word32)WEBRTC_SPL_MUL_16_16_RSFT((kAllPassCoefsQ13[1]), tmp16_2, 12);
+    }
+    filter_state[0] = tmp32_1;
+    filter_state[1] = tmp32_2;
+}
+
+WebRtc_Word16 WebRtcVad_FindMinimum(VadInstT* inst,
+                                    WebRtc_Word16 x,
+                                    int n)
+{
+    int i, j, k, II = -1, offset;
+    WebRtc_Word16 meanV, alpha;
+    WebRtc_Word32 tmp32, tmp32_1;
+    WebRtc_Word16 *valptr, *idxptr, *p1, *p2, *p3;
+
+    // Offset to beginning of the 16 minimum values in memory
+    offset = WEBRTC_SPL_LSHIFT_W16(n, 4);
+
+    // Pointer to memory for the 16 minimum values and the age of each value
+    idxptr = &inst->index_vector[offset];
+    valptr = &inst->low_value_vector[offset];
+
+    // Each value in low_value_vector is getting 1 loop older.
+    // Update age of each value in indexVal, and remove old values.
+    for (i = 0; i < 16; i++)
+    {
+        p3 = idxptr + i;
+        if (*p3 != 100)
+        {
+            *p3 += 1;
+        } else
+        {
+            p1 = valptr + i + 1;
+            p2 = p3 + 1;
+            for (j = i; j < 16; j++)
+            {
+                *(valptr + j) = *p1++;
+                *(idxptr + j) = *p2++;
+            }
+            *(idxptr + 15) = 101;
+            *(valptr + 15) = 10000;
+        }
+    }
+
+    // Check if x smaller than any of the values in low_value_vector.
+    // If so, find position.
+    if (x < *(valptr + 7))
+    {
+        if (x < *(valptr + 3))
+        {
+            if (x < *(valptr + 1))
+            {
+                if (x < *valptr)
+                {
+                    II = 0;
+                } else
+                {
+                    II = 1;
+                }
+            } else if (x < *(valptr + 2))
+            {
+                II = 2;
+            } else
+            {
+                II = 3;
+            }
+        } else if (x < *(valptr + 5))
+        {
+            if (x < *(valptr + 4))
+            {
+                II = 4;
+            } else
+            {
+                II = 5;
+            }
+        } else if (x < *(valptr + 6))
+        {
+            II = 6;
+        } else
+        {
+            II = 7;
+        }
+    } else if (x < *(valptr + 15))
+    {
+        if (x < *(valptr + 11))
+        {
+            if (x < *(valptr + 9))
+            {
+                if (x < *(valptr + 8))
+                {
+                    II = 8;
+                } else
+                {
+                    II = 9;
+                }
+            } else if (x < *(valptr + 10))
+            {
+                II = 10;
+            } else
+            {
+                II = 11;
+            }
+        } else if (x < *(valptr + 13))
+        {
+            if (x < *(valptr + 12))
+            {
+                II = 12;
+            } else
+            {
+                II = 13;
+            }
+        } else if (x < *(valptr + 14))
+        {
+            II = 14;
+        } else
+        {
+            II = 15;
+        }
+    }
+
+    // Put new min value on right position and shift bigger values up
+    if (II > -1)
+    {
+        for (i = 15; i > II; i--)
+        {
+            k = i - 1;
+            *(valptr + i) = *(valptr + k);
+            *(idxptr + i) = *(idxptr + k);
+        }
+        *(valptr + II) = x;
+        *(idxptr + II) = 1;
+    }
+
+    meanV = 0;
+    if ((inst->frame_counter) > 4)
+    {
+        j = 5;
+    } else
+    {
+        j = inst->frame_counter;
+    }
+
+    if (j > 2)
+    {
+        meanV = *(valptr + 2);
+    } else if (j > 0)
+    {
+        meanV = *valptr;
+    } else
+    {
+        meanV = 1600;
+    }
+
+    if (inst->frame_counter > 0)
+    {
+        if (meanV < inst->mean_value[n])
+        {
+            alpha = (WebRtc_Word16)ALPHA1; // 0.2 in Q15
+        } else
+        {
+            alpha = (WebRtc_Word16)ALPHA2; // 0.99 in Q15
+        }
+    } else
+    {
+        alpha = 0;
+    }
+
+    tmp32 = WEBRTC_SPL_MUL_16_16((alpha+1), inst->mean_value[n]);
+    tmp32_1 = WEBRTC_SPL_MUL_16_16(WEBRTC_SPL_WORD16_MAX - alpha, meanV);
+    tmp32 += tmp32_1;
+    tmp32 += 16384;
+    inst->mean_value[n] = (WebRtc_Word16)WEBRTC_SPL_RSHIFT_W32(tmp32, 15);
+
+    return inst->mean_value[n];
+}

src/common_audio/vad/main/source/vad_sp.h

@@ -0,0 +1,60 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This header file includes the VAD internal calls for Downsampling and FindMinimum.
+ * Specific function calls are given below.
+ */
+
+#ifndef WEBRTC_VAD_SP_H_
+#define WEBRTC_VAD_SP_H_
+
+#include "vad_core.h"
+
+/****************************************************************************
+ * WebRtcVad_Downsampling(...)
+ *
+ * Downsamples the signal a factor 2, eg. 32->16 or 16->8
+ *
+ * Input:
+ *      - signal_in     : Input signal
+ *      - in_length     : Length of input signal in samples
+ *
+ * Input & Output:
+ *      - filter_state  : Filter state for first all-pass filters
+ *
+ * Output:
+ *      - signal_out    : Downsampled signal (of length len/2)
+ */
+void WebRtcVad_Downsampling(WebRtc_Word16* signal_in,
+                            WebRtc_Word16* signal_out,
+                            WebRtc_Word32* filter_state,
+                            int in_length);
+
+/****************************************************************************
+ * WebRtcVad_FindMinimum(...)
+ *
+ * Find the five lowest values of x in 100 frames long window. Return a mean
+ * value of these five values.
+ *
+ * Input:
+ *      - feature_value : Feature value
+ *      - channel       : Channel number
+ *
+ * Input & Output:
+ *      - inst          : State information
+ *
+ * Output:
+ *      return value    : Weighted minimum value for a moving window.
+ */
+WebRtc_Word16 WebRtcVad_FindMinimum(VadInstT* inst, WebRtc_Word16 feature_value, int channel);
+
+#endif // WEBRTC_VAD_SP_H_

src/common_audio/vad/main/source/webrtc_vad.c

@@ -0,0 +1,197 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file includes the VAD API calls. For a specific function call description,
+ * see webrtc_vad.h
+ */
+
+#include <stdlib.h>
+#include <string.h>
+
+#include "webrtc_vad.h"
+#include "vad_core.h"
+
+static const int kInitCheck = 42;
+
+WebRtc_Word16 WebRtcVad_get_version(char *version, size_t size_bytes)
+{
+    const char my_version[] = "VAD 1.2.0";
+
+    if (version == NULL)
+    {
+        return -1;
+    }
+
+    if (size_bytes < sizeof(my_version))
+    {
+        return -1;
+    }
+
+    memcpy(version, my_version, sizeof(my_version));
+    return 0;
+}
+
+WebRtc_Word16 WebRtcVad_AssignSize(int *size_in_bytes)
+{
+    *size_in_bytes = sizeof(VadInstT) * 2 / sizeof(WebRtc_Word16);
+    return 0;
+}
+
+WebRtc_Word16 WebRtcVad_Assign(VadInst **vad_inst, void *vad_inst_addr)
+{
+
+    if (vad_inst == NULL)
+    {
+        return -1;
+    }
+
+    if (vad_inst_addr != NULL)
+    {
+        *vad_inst = (VadInst*)vad_inst_addr;
+        return 0;
+    } else
+    {
+        return -1;
+    }
+}
+
+WebRtc_Word16 WebRtcVad_Create(VadInst **vad_inst)
+{
+
+    VadInstT *vad_ptr = NULL;
+
+    if (vad_inst == NULL)
+    {
+        return -1;
+    }
+
+    *vad_inst = NULL;
+
+    vad_ptr = (VadInstT *)malloc(sizeof(VadInstT));
+    *vad_inst = (VadInst *)vad_ptr;
+
+    if (vad_ptr == NULL)
+    {
+        return -1;
+    }
+
+    vad_ptr->init_flag = 0;
+
+    return 0;
+}
+
+WebRtc_Word16 WebRtcVad_Free(VadInst *vad_inst)
+{
+
+    if (vad_inst == NULL)
+    {
+        return -1;
+    }
+
+    free(vad_inst);
+    return 0;
+}
+
+WebRtc_Word16 WebRtcVad_Init(VadInst *vad_inst)
+{
+    short mode = 0; // Default high quality
+
+    if (vad_inst == NULL)
+    {
+        return -1;
+    }
+
+    return WebRtcVad_InitCore((VadInstT*)vad_inst, mode);
+}
+
+WebRtc_Word16 WebRtcVad_set_mode(VadInst *vad_inst, WebRtc_Word16 mode)
+{
+    VadInstT* vad_ptr;
+
+    if (vad_inst == NULL)
+    {
+        return -1;
+    }
+
+    vad_ptr = (VadInstT*)vad_inst;
+    if (vad_ptr->init_flag != kInitCheck)
+    {
+        return -1;
+    }
+
+    return WebRtcVad_set_mode_core((VadInstT*)vad_inst, mode);
+}
+
+WebRtc_Word16 WebRtcVad_Process(VadInst *vad_inst,
+                                WebRtc_Word16 fs,
+                                WebRtc_Word16 *speech_frame,
+                                WebRtc_Word16 frame_length)
+{
+    WebRtc_Word16 vad;
+    VadInstT* vad_ptr;
+
+    if (vad_inst == NULL)
+    {
+        return -1;
+    }
+
+    vad_ptr = (VadInstT*)vad_inst;
+    if (vad_ptr->init_flag != kInitCheck)
+    {
+        return -1;
+    }
+
+    if (speech_frame == NULL)
+    {
+        return -1;
+    }
+
+    if (fs == 32000)
+    {
+        if ((frame_length != 320) && (frame_length != 640) && (frame_length != 960))
+        {
+            return -1;
+        }
+        vad = WebRtcVad_CalcVad32khz((VadInstT*)vad_inst, speech_frame, frame_length);
+
+    } else if (fs == 16000)
+    {
+        if ((frame_length != 160) && (frame_length != 320) && (frame_length != 480))
+        {
+            return -1;
+        }
+        vad = WebRtcVad_CalcVad16khz((VadInstT*)vad_inst, speech_frame, frame_length);
+
+    } else if (fs == 8000)
+    {
+        if ((frame_length != 80) && (frame_length != 160) && (frame_length != 240))
+        {
+            return -1;
+        }
+        vad = WebRtcVad_CalcVad8khz((VadInstT*)vad_inst, speech_frame, frame_length);
+
+    } else
+    {
+        return -1; // Not a supported sampling frequency
+    }
+
+    if (vad > 0)
+    {
+        return 1;
+    } else if (vad == 0)
+    {
+        return 0;
+    } else
+    {
+        return -1;
+    }
+}

src/common_audio/vad/main/test/unit_test/unit_test.cc

@@ -0,0 +1,123 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This file includes the implementation of the VAD unit tests.
+ */
+
+#include <cstring>
+#include "unit_test.h"
+#include "webrtc_vad.h"
+
+
+class VadEnvironment : public ::testing::Environment {
+ public:
+  virtual void SetUp() {
+  }
+
+  virtual void TearDown() {
+  }
+};
+
+VadTest::VadTest()
+{
+}
+
+void VadTest::SetUp() {
+}
+
+void VadTest::TearDown() {
+}
+
+TEST_F(VadTest, ApiTest) {
+    VadInst *vad_inst;
+    int i, j, k;
+    short zeros[960];
+    short speech[960];
+    char version[32];
+
+    // Valid test cases
+    int fs[3] = {8000, 16000, 32000};
+    int nMode[4] = {0, 1, 2, 3};
+    int framelen[3][3] = {{80, 160, 240},
+    {160, 320, 480}, {320, 640, 960}} ;
+    int vad_counter = 0;
+
+    memset(zeros, 0, sizeof(short) * 960);
+    memset(speech, 1, sizeof(short) * 960);
+    speech[13] = 1374;
+    speech[73] = -3747;
+
+
+
+    // WebRtcVad_get_version()
+    WebRtcVad_get_version(version);
+    //printf("API Test for %s\n", version);
+
+    // Null instance tests
+    EXPECT_EQ(-1, WebRtcVad_Create(NULL));
+    EXPECT_EQ(-1, WebRtcVad_Init(NULL));
+    EXPECT_EQ(-1, WebRtcVad_Assign(NULL, NULL));
+    EXPECT_EQ(-1, WebRtcVad_Free(NULL));
+    EXPECT_EQ(-1, WebRtcVad_set_mode(NULL, nMode[0]));
+    EXPECT_EQ(-1, WebRtcVad_Process(NULL, fs[0], speech,  framelen[0][0]));
+
+
+    EXPECT_EQ(WebRtcVad_Create(&vad_inst), 0);
+
+    // Not initialized tests
+    EXPECT_EQ(-1, WebRtcVad_Process(vad_inst, fs[0], speech,  framelen[0][0]));
+    EXPECT_EQ(-1, WebRtcVad_set_mode(vad_inst, nMode[0]));
+
+    // WebRtcVad_Init() tests
+    EXPECT_EQ(WebRtcVad_Init(vad_inst), 0);
+
+    // WebRtcVad_set_mode() tests
+    EXPECT_EQ(-1, WebRtcVad_set_mode(vad_inst, -1));
+    EXPECT_EQ(-1, WebRtcVad_set_mode(vad_inst, 4));
+
+    for (i = 0; i < sizeof(nMode)/sizeof(nMode[0]); i++) {
+        EXPECT_EQ(WebRtcVad_set_mode(vad_inst, nMode[i]), 0);
+    }
+
+    // WebRtcVad_Process() tests
+    EXPECT_EQ(-1, WebRtcVad_Process(vad_inst, fs[0], NULL,  framelen[0][0]));
+    EXPECT_EQ(-1, WebRtcVad_Process(vad_inst, 12000, speech,  framelen[0][0]));
+    EXPECT_EQ(-1, WebRtcVad_Process(vad_inst, fs[0], speech,  framelen[1][1]));
+    EXPECT_EQ(WebRtcVad_Process(vad_inst, fs[0], zeros,  framelen[0][0]), 0);
+    for (i = 0; i < sizeof(fs)/sizeof(fs[0]); i++) {
+        for (j = 0; j < sizeof(framelen[0])/sizeof(framelen[0][0]); j++) {
+            for (k = 0; k < sizeof(nMode)/sizeof(nMode[0]); k++) {
+                EXPECT_EQ(WebRtcVad_set_mode(vad_inst, nMode[k]), 0);
+//                printf("%d\n", WebRtcVad_Process(vad_inst, fs[i], speech,  framelen[i][j]));
+                if (vad_counter < 9)
+                {
+                    EXPECT_EQ(WebRtcVad_Process(vad_inst, fs[i], speech,  framelen[i][j]), 1);
+                } else
+                {
+                    EXPECT_EQ(WebRtcVad_Process(vad_inst, fs[i], speech,  framelen[i][j]), 0);
+                }
+                vad_counter++;
+            }
+        }
+    }
+
+    EXPECT_EQ(0, WebRtcVad_Free(vad_inst));
+
+}
+
+int main(int argc, char** argv) {
+  ::testing::InitGoogleTest(&argc, argv);
+  VadEnvironment* env = new VadEnvironment;
+  ::testing::AddGlobalTestEnvironment(env);
+
+  return RUN_ALL_TESTS();
+}

src/common_audio/vad/main/test/unit_test/unit_test.h

@@ -0,0 +1,28 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+
+/*
+ * This header file includes the declaration of the VAD unit test.
+ */
+
+#ifndef WEBRTC_VAD_UNIT_TEST_H_
+#define WEBRTC_VAD_UNIT_TEST_H_
+
+#include <gtest/gtest.h>
+
+class VadTest : public ::testing::Test {
+ protected:
+  VadTest();
+  virtual void SetUp();
+  virtual void TearDown();
+};
+
+#endif  // WEBRTC_VAD_UNIT_TEST_H_

src/common_types.h

@@ -0,0 +1,595 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef WEBRTC_COMMON_TYPES_H
+#define WEBRTC_COMMON_TYPES_H
+
+#include "typedefs.h"
+
+#ifdef WEBRTC_EXPORT
+    #define WEBRTC_DLLEXPORT _declspec(dllexport)
+#elif WEBRTC_DLL
+    #define WEBRTC_DLLEXPORT _declspec(dllimport)
+#else
+    #define WEBRTC_DLLEXPORT
+#endif
+
+#ifndef NULL
+    #define NULL 0
+#endif
+
+namespace webrtc {
+
+class InStream
+{
+public:
+    virtual int Read(void *buf,int len) = 0;
+    virtual int Rewind() {return -1;}
+    virtual ~InStream() {}
+protected:
+    InStream() {}
+};
+
+class OutStream
+{
+public:
+    virtual bool Write(const void *buf,int len) = 0;
+    virtual int Rewind() {return -1;}
+    virtual ~OutStream() {}
+protected:
+    OutStream() {}
+};
+
+enum TraceModule
+{
+    // not a module, triggered from the engine code
+    kTraceVoice              = 0x0001,
+    // not a module, triggered from the engine code
+    kTraceVideo              = 0x0002,
+    // not a module, triggered from the utility code
+    kTraceUtility            = 0x0003,
+    kTraceRtpRtcp            = 0x0004,
+    kTraceTransport          = 0x0005,
+    kTraceSrtp               = 0x0006,
+    kTraceAudioCoding        = 0x0007,
+    kTraceAudioMixerServer   = 0x0008,
+    kTraceAudioMixerClient   = 0x0009,
+    kTraceFile               = 0x000a,
+    kTraceAudioProcessing    = 0x000b,
+    kTraceVideoCoding        = 0x0010,
+    kTraceVideoMixer         = 0x0011,
+    kTraceAudioDevice        = 0x0012,
+    kTraceVideoRenderer      = 0x0014,
+    kTraceVideoCapture       = 0x0015,
+    kTraceVideoPreocessing   = 0x0016
+};
+
+enum TraceLevel
+{
+    kTraceNone               = 0x0000,    // no trace
+    kTraceStateInfo          = 0x0001,
+    kTraceWarning            = 0x0002,
+    kTraceError              = 0x0004,
+    kTraceCritical           = 0x0008,
+    kTraceApiCall            = 0x0010,
+    kTraceDefault            = 0x00ff,
+
+    kTraceModuleCall         = 0x0020,
+    kTraceMemory             = 0x0100,   // memory info
+    kTraceTimer              = 0x0200,   // timing info
+    kTraceStream             = 0x0400,   // "continuous" stream of data
+
+    // used for debug purposes
+    kTraceDebug              = 0x0800,  // debug
+    kTraceInfo               = 0x1000,  // debug info
+
+    kTraceAll                = 0xffff
+};
+
+// External Trace API
+class TraceCallback
+{
+public:
+    virtual void Print(const TraceLevel level,
+                       const char *traceString,
+                       const int length) = 0;
+protected:
+    virtual ~TraceCallback() {}
+    TraceCallback() {}
+};
+
+
+enum FileFormats
+{
+    kFileFormatWavFile        = 1,
+    kFileFormatCompressedFile = 2,
+    kFileFormatAviFile        = 3,
+    kFileFormatPreencodedFile = 4,
+    kFileFormatPcm16kHzFile   = 7,
+    kFileFormatPcm8kHzFile    = 8,
+    kFileFormatPcm32kHzFile   = 9
+};
+
+
+enum ProcessingTypes
+{
+    kPlaybackPerChannel = 0,
+    kPlaybackAllChannelsMixed,
+    kRecordingPerChannel,
+    kRecordingAllChannelsMixed
+};
+
+// Encryption enums
+enum CipherTypes
+{
+    kCipherNull               = 0,
+    kCipherAes128CounterMode  = 1
+};
+
+enum AuthenticationTypes
+{
+    kAuthNull       = 0,
+    kAuthHmacSha1   = 3
+};
+
+enum SecurityLevels
+{
+    kNoProtection                    = 0,
+    kEncryption                      = 1,
+    kAuthentication                  = 2,
+    kEncryptionAndAuthentication     = 3
+};
+
+class Encryption
+{
+public:
+    virtual void encrypt(
+        int channel_no,
+        unsigned char* in_data,
+        unsigned char* out_data,
+        int bytes_in,
+        int* bytes_out) = 0;
+
+    virtual void decrypt(
+        int channel_no,
+        unsigned char* in_data,
+        unsigned char* out_data,
+        int bytes_in,
+        int* bytes_out) = 0;
+
+    virtual void encrypt_rtcp(
+        int channel_no,
+        unsigned char* in_data,
+        unsigned char* out_data,
+        int bytes_in,
+        int* bytes_out) = 0;
+
+    virtual void decrypt_rtcp(
+        int channel_no,
+        unsigned char* in_data,
+        unsigned char* out_data,
+        int bytes_in,
+        int* bytes_out) = 0;
+
+protected:
+    virtual ~Encryption() {}
+    Encryption() {}
+};
+
+// External transport callback interface
+class Transport
+{
+public:
+    virtual int SendPacket(int channel, const void *data, int len) = 0;
+    virtual int SendRTCPPacket(int channel, const void *data, int len) = 0;
+
+protected:
+    virtual ~Transport() {}
+    Transport() {}
+};
+
+// ==================================================================
+// Voice specific types
+// ==================================================================
+
+// Each codec supported can be described by this structure.
+struct CodecInst
+{
+    int pltype;
+    char plname[32];
+    int plfreq;
+    int pacsize;
+    int channels;
+    int rate;
+};
+
+enum FrameType
+{
+    kFrameEmpty            = 0,
+    kAudioFrameSpeech      = 1,
+    kAudioFrameCN          = 2,
+    kVideoFrameKey         = 3,    // independent frame
+    kVideoFrameDelta       = 4,    // depends on the previus frame
+    kVideoFrameGolden      = 5,    // depends on a old known previus frame
+    kVideoFrameAltRef      = 6
+};
+
+// RTP
+enum {kRtpCsrcSize = 15}; // RFC 3550 page 13
+
+enum RTPDirections
+{
+    kRtpIncoming = 0,
+    kRtpOutgoing
+};
+
+enum PayloadFrequencies
+{
+    kFreq8000Hz = 8000,
+    kFreq16000Hz = 16000,
+    kFreq32000Hz = 32000
+};
+
+enum VadModes                 // degree of bandwidth reduction
+{
+    kVadConventional = 0,      // lowest reduction
+    kVadAggressiveLow,
+    kVadAggressiveMid,
+    kVadAggressiveHigh         // highest reduction
+};
+
+struct NetworkStatistics           // NETEQ statistics
+{
+    // current jitter buffer size in ms
+    WebRtc_UWord16 currentBufferSize;
+    // preferred (optimal) buffer size in ms
+    WebRtc_UWord16 preferredBufferSize;
+    // loss rate (network + late) in percent (in Q14)
+    WebRtc_UWord16 currentPacketLossRate;
+    // late loss rate in percent (in Q14)
+    WebRtc_UWord16 currentDiscardRate;
+    // fraction (of original stream) of synthesized speech inserted through
+    // expansion (in Q14)
+    WebRtc_UWord16 currentExpandRate;
+    // fraction of synthesized speech inserted through pre-emptive expansion
+    // (in Q14)
+    WebRtc_UWord16 currentPreemptiveRate;
+    // fraction of data removed through acceleration (in Q14)
+    WebRtc_UWord16 currentAccelerateRate;
+};
+
+struct JitterStatistics
+{
+    // smallest Jitter Buffer size during call in ms
+    WebRtc_UWord32 jbMinSize;
+    // largest Jitter Buffer size during call in ms
+    WebRtc_UWord32 jbMaxSize;
+    // the average JB size, measured over time - ms
+    WebRtc_UWord32 jbAvgSize;
+    // number of times the Jitter Buffer changed (using Accelerate or
+    // Pre-emptive Expand)
+    WebRtc_UWord32 jbChangeCount;
+    // amount (in ms) of audio data received late
+    WebRtc_UWord32 lateLossMs;
+    // milliseconds removed to reduce jitter buffer size
+    WebRtc_UWord32 accelerateMs;
+    // milliseconds discarded through buffer flushing
+    WebRtc_UWord32 flushedMs;
+    // milliseconds of generated silence
+    WebRtc_UWord32 generatedSilentMs;
+    // milliseconds of synthetic audio data (non-background noise)
+    WebRtc_UWord32 interpolatedVoiceMs;
+    // milliseconds of synthetic audio data (background noise level)
+    WebRtc_UWord32 interpolatedSilentMs;
+    // count of tiny expansions in output audio
+    WebRtc_UWord32 countExpandMoreThan120ms;
+    // count of small expansions in output audio
+    WebRtc_UWord32 countExpandMoreThan250ms;
+    // count of medium expansions in output audio
+    WebRtc_UWord32 countExpandMoreThan500ms;
+    // count of long expansions in output audio
+    WebRtc_UWord32 countExpandMoreThan2000ms;
+    // duration of longest audio drop-out
+    WebRtc_UWord32 longestExpandDurationMs;
+    // count of times we got small network outage (inter-arrival time in
+    // [500, 1000) ms)
+    WebRtc_UWord32 countIAT500ms;
+    // count of times we got medium network outage (inter-arrival time in
+    // [1000, 2000) ms)
+    WebRtc_UWord32 countIAT1000ms;
+    // count of times we got large network outage (inter-arrival time >=
+    // 2000 ms)
+    WebRtc_UWord32 countIAT2000ms;
+    // longest packet inter-arrival time in ms
+    WebRtc_UWord32 longestIATms;
+    // min time incoming Packet "waited" to be played
+    WebRtc_UWord32 minPacketDelayMs;
+    // max time incoming Packet "waited" to be played
+    WebRtc_UWord32 maxPacketDelayMs;
+    // avg time incoming Packet "waited" to be played
+    WebRtc_UWord32 avgPacketDelayMs;
+};
+
+typedef struct
+{
+    int min;              // minumum
+    int max;              // maximum
+    int average;          // average
+} StatVal;
+
+typedef struct           // All levels are reported in dBm0
+{
+    StatVal speech_rx;   // long-term speech levels on receiving side
+    StatVal speech_tx;   // long-term speech levels on transmitting side
+    StatVal noise_rx;    // long-term noise/silence levels on receiving side
+    StatVal noise_tx;    // long-term noise/silence levels on transmitting side
+} LevelStatistics;
+
+typedef struct        // All levels are reported in dB
+{
+    StatVal erl;      // Echo Return Loss
+    StatVal erle;     // Echo Return Loss Enhancement
+    StatVal rerl;     // RERL = ERL + ERLE
+    // Echo suppression inside EC at the point just before its NLP
+    StatVal a_nlp;
+} EchoStatistics;
+
+enum TelephoneEventDetectionMethods
+{
+    kInBand = 0,
+    kOutOfBand = 1,
+    kInAndOutOfBand = 2
+};
+
+enum NsModes    // type of Noise Suppression
+{
+    kNsUnchanged = 0,   // previously set mode
+    kNsDefault,         // platform default
+    kNsConference,      // conferencing default
+    kNsLowSuppression,  // lowest suppression
+    kNsModerateSuppression,
+    kNsHighSuppression,
+    kNsVeryHighSuppression,     // highest suppression
+};
+
+enum AgcModes                  // type of Automatic Gain Control
+{
+    kAgcUnchanged = 0,        // previously set mode
+    kAgcDefault,              // platform default
+    // adaptive mode for use when analog volume control exists (e.g. for
+    // PC softphone)
+    kAgcAdaptiveAnalog,
+    // scaling takes place in the digital domain (e.g. for conference servers
+    // and embedded devices)
+    kAgcAdaptiveDigital,
+    // can be used on embedded devices where the the capture signal is level
+    // is predictable
+    kAgcFixedDigital
+};
+
+// EC modes
+enum EcModes                   // type of Echo Control
+{
+    kEcUnchanged = 0,          // previously set mode
+    kEcDefault,                // platform default
+    kEcConference,             // conferencing default (aggressive AEC)
+    kEcAec,                    // Acoustic Echo Cancellation
+    kEcAecm,                   // AEC mobile
+};
+
+// AECM modes
+enum AecmModes                 // mode of AECM
+{
+    kAecmQuietEarpieceOrHeadset = 0,
+                               // Quiet earpiece or headset use
+    kAecmEarpiece,             // most earpiece use
+    kAecmLoudEarpiece,         // Loud earpiece or quiet speakerphone use
+    kAecmSpeakerphone,         // most speakerphone use (default)
+    kAecmLoudSpeakerphone      // Loud speakerphone
+};
+
+// AGC configuration
+typedef struct
+{
+    unsigned short targetLeveldBOv;
+    unsigned short digitalCompressionGaindB;
+    bool           limiterEnable;
+} AgcConfig;                  // AGC configuration parameters
+
+enum StereoChannel
+{
+    kStereoLeft = 0,
+    kStereoRight,
+    kStereoBoth
+};
+
+// Audio device layers
+enum AudioLayers
+{
+    kAudioPlatformDefault = 0,
+    kAudioWindowsWave = 1,
+    kAudioWindowsCore = 2,
+    kAudioLinuxAlsa = 3,
+    kAudioLinuxPulse = 4
+};
+
+enum NetEqModes             // NetEQ playout configurations
+{
+    // Optimized trade-off between low delay and jitter robustness for two-way
+    // communication.
+    kNetEqDefault = 0,
+    // Improved jitter robustness at the cost of increased delay. Can be
+    // used in one-way communication.
+    kNetEqStreaming = 1,
+    // Optimzed for decodability of fax signals rather than for perceived audio
+    // quality.
+    kNetEqFax = 2,
+};
+
+enum NetEqBgnModes          // NetEQ Background Noise (BGN) configurations
+{
+    // BGN is always on and will be generated when the incoming RTP stream
+    // stops (default).
+    kBgnOn = 0,
+    // The BGN is faded to zero (complete silence) after a few seconds.
+    kBgnFade = 1,
+    // BGN is not used at all. Silence is produced after speech extrapolation
+    // has faded.
+    kBgnOff = 2,
+};
+
+enum OnHoldModes            // On Hold direction
+{
+    kHoldSendAndPlay = 0,    // Put both sending and playing in on-hold state.
+    kHoldSendOnly,           // Put only sending in on-hold state.
+    kHoldPlayOnly            // Put only playing in on-hold state.
+};
+
+enum AmrMode
+{
+    kRfc3267BwEfficient = 0,
+    kRfc3267OctetAligned = 1,
+    kRfc3267FileStorage = 2,
+};
+
+// ==================================================================
+// Video specific types
+// ==================================================================
+
+// Raw video types
+enum RawVideoType
+{
+    kVideoI420     = 0,
+    kVideoYV12     = 1,
+    kVideoYUY2     = 2,
+    kVideoUYVY     = 3,
+    kVideoIYUV     = 4,
+    kVideoARGB     = 5,
+    kVideoRGB24    = 6,
+    kVideoRGB565   = 7,
+    kVideoARGB4444 = 8,
+    kVideoARGB1555 = 9,
+    kVideoMJPEG    = 10,
+    kVideoNV12     = 11,
+    kVideoNV21     = 12,
+    kVideoUnknown  = 99
+};
+
+// Video codec
+enum { kConfigParameterSize = 128};
+enum { kPayloadNameSize = 32};
+
+// H.263 specific
+struct VideoCodecH263
+{
+    char quality;
+};
+
+// H.264 specific
+enum H264Packetization
+{
+    kH264SingleMode         = 0,
+    kH264NonInterleavedMode = 1
+};
+
+enum VideoCodecComplexity
+{
+    kComplexityNormal = 0,
+    kComplexityHigh    = 1,
+    kComplexityHigher  = 2,
+    kComplexityMax     = 3
+};
+
+enum VideoCodecProfile
+{
+    kProfileBase = 0x00,
+    kProfileMain = 0x01
+};
+
+struct VideoCodecH264
+{
+    H264Packetization          packetization;
+    VideoCodecComplexity       complexity;
+    VideoCodecProfile          profile;
+    char                       level;
+    char                       quality;
+
+    bool                       useFMO;
+
+    unsigned char              configParameters[kConfigParameterSize];
+    unsigned char              configParametersSize;
+};
+
+// VP8 specific
+struct VideoCodecVP8
+{
+    bool                       pictureLossIndicationOn;
+    bool                       feedbackModeOn;
+    VideoCodecComplexity       complexity;
+};
+
+// MPEG-4 specific
+struct VideoCodecMPEG4
+{
+    unsigned char   configParameters[kConfigParameterSize];
+    unsigned char   configParametersSize;
+    char            level;
+};
+
+// Unknown specific
+struct VideoCodecGeneric
+{
+};
+
+// Video codec types
+enum VideoCodecType
+{
+    kVideoCodecH263,
+    kVideoCodecH264,
+    kVideoCodecVP8,
+    kVideoCodecMPEG4,
+    kVideoCodecI420,
+    kVideoCodecRED,
+    kVideoCodecULPFEC,
+    kVideoCodecUnknown
+};
+
+union VideoCodecUnion
+{
+    VideoCodecH263      H263;
+    VideoCodecH264      H264;
+    VideoCodecVP8       VP8;
+    VideoCodecMPEG4     MPEG4;
+    VideoCodecGeneric   Generic;
+};
+
+// Common video codec properties
+struct VideoCodec
+{
+    VideoCodecType      codecType;
+    char                plName[kPayloadNameSize];
+    unsigned char       plType;
+
+    unsigned short      width;
+    unsigned short      height;
+
+    unsigned int        startBitrate;
+    unsigned int        maxBitrate;
+    unsigned int        minBitrate;
+    unsigned char       maxFramerate;
+
+    VideoCodecUnion     codecSpecific;
+
+    unsigned int        qpMax;
+};
+
+}  // namespace webrtc
+
+#endif  // WEBRTC_COMMON_TYPES_H

src/modules/audio_processing/OWNERS

@@ -0,0 +1,2 @@
+andrew@webrtc.org
+bjornv@webrtc.org

src/modules/audio_processing/aec/main/interface/echo_cancellation.h

@@ -0,0 +1,260 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_INTERFACE_ECHO_CANCELLATION_H_
+#define WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_INTERFACE_ECHO_CANCELLATION_H_
+
+#include "typedefs.h"
+
+// Errors
+#define AEC_UNSPECIFIED_ERROR           12000
+#define AEC_UNSUPPORTED_FUNCTION_ERROR  12001
+#define AEC_UNINITIALIZED_ERROR         12002
+#define AEC_NULL_POINTER_ERROR          12003
+#define AEC_BAD_PARAMETER_ERROR         12004
+
+// Warnings
+#define AEC_BAD_PARAMETER_WARNING       12050
+
+enum {
+    kAecNlpConservative = 0,
+    kAecNlpModerate,
+    kAecNlpAggressive
+};
+
+enum {
+    kAecFalse = 0,
+    kAecTrue
+};
+
+typedef struct {
+    WebRtc_Word16 nlpMode;        // default kAecNlpModerate
+    WebRtc_Word16 skewMode;       // default kAecFalse
+    WebRtc_Word16 metricsMode;    // default kAecFalse
+    //float realSkew;
+} AecConfig;
+
+typedef struct {
+    WebRtc_Word16 instant;
+    WebRtc_Word16 average;
+    WebRtc_Word16 max;
+    WebRtc_Word16 min;
+} AecLevel;
+
+typedef struct {
+    AecLevel rerl;
+    AecLevel erl;
+    AecLevel erle;
+    AecLevel aNlp;
+} AecMetrics;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/*
+ * Allocates the memory needed by the AEC. The memory needs to be initialized
+ * separately using the WebRtcAec_Init() function.
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void **aecInst               Pointer to the AEC instance to be created
+ *                              and initilized
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word32 return          0: OK
+ *                              -1: error
+ */
+WebRtc_Word32 WebRtcAec_Create(void **aecInst);
+
+/*
+ * This function releases the memory allocated by WebRtcAec_Create().
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void         *aecInst        Pointer to the AEC instance
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word32  return         0: OK
+ *                              -1: error
+ */
+WebRtc_Word32 WebRtcAec_Free(void *aecInst);
+
+/*
+ * Initializes an AEC instance.
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void           *aecInst      Pointer to the AEC instance
+ * WebRtc_Word32  sampFreq      Sampling frequency of data
+ * WebRtc_Word32  scSampFreq    Soundcard sampling frequency
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word32 return          0: OK
+ *                              -1: error
+ */
+WebRtc_Word32 WebRtcAec_Init(void *aecInst,
+                             WebRtc_Word32 sampFreq,
+                             WebRtc_Word32 scSampFreq);
+
+/*
+ * Inserts an 80 or 160 sample block of data into the farend buffer.
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void           *aecInst      Pointer to the AEC instance
+ * WebRtc_Word16  *farend       In buffer containing one frame of
+ *                              farend signal for L band
+ * WebRtc_Word16  nrOfSamples   Number of samples in farend buffer
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word32  return         0: OK
+ *                              -1: error
+ */
+WebRtc_Word32 WebRtcAec_BufferFarend(void *aecInst,
+                                     const WebRtc_Word16 *farend,
+                                     WebRtc_Word16 nrOfSamples);
+
+/*
+ * Runs the echo canceller on an 80 or 160 sample blocks of data.
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void          *aecInst       Pointer to the AEC instance
+ * WebRtc_Word16 *nearend       In buffer containing one frame of
+ *                              nearend+echo signal for L band
+ * WebRtc_Word16 *nearendH      In buffer containing one frame of
+ *                              nearend+echo signal for H band
+ * WebRtc_Word16 nrOfSamples    Number of samples in nearend buffer
+ * WebRtc_Word16 msInSndCardBuf Delay estimate for sound card and
+ *                              system buffers
+ * WebRtc_Word16 skew           Difference between number of samples played
+ *                              and recorded at the soundcard (for clock skew
+ *                              compensation)
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word16  *out          Out buffer, one frame of processed nearend
+ *                              for L band
+ * WebRtc_Word16  *outH         Out buffer, one frame of processed nearend
+ *                              for H band
+ * WebRtc_Word32  return         0: OK
+ *                              -1: error
+ */
+WebRtc_Word32 WebRtcAec_Process(void *aecInst,
+                                const WebRtc_Word16 *nearend,
+                                const WebRtc_Word16 *nearendH,
+                                WebRtc_Word16 *out,
+                                WebRtc_Word16 *outH,
+                                WebRtc_Word16 nrOfSamples,
+                                WebRtc_Word16 msInSndCardBuf,
+                                WebRtc_Word32 skew);
+
+/*
+ * This function enables the user to set certain parameters on-the-fly.
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void           *aecInst      Pointer to the AEC instance
+ * AecConfig      config        Config instance that contains all
+ *                              properties to be set
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word32  return         0: OK
+ *                              -1: error
+ */
+WebRtc_Word32 WebRtcAec_set_config(void *aecInst, AecConfig config);
+
+/*
+ * Gets the on-the-fly paramters.
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void           *aecInst      Pointer to the AEC instance
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * AecConfig      *config       Pointer to the config instance that
+ *                              all properties will be written to
+ * WebRtc_Word32  return         0: OK
+ *                              -1: error
+ */
+WebRtc_Word32 WebRtcAec_get_config(void *aecInst, AecConfig *config);
+
+/*
+ * Gets the current echo status of the nearend signal.
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void           *aecInst      Pointer to the AEC instance
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word16  *status       0: Almost certainly nearend single-talk
+ *                              1: Might not be neared single-talk
+ * WebRtc_Word32  return         0: OK
+ *                              -1: error
+ */
+WebRtc_Word32 WebRtcAec_get_echo_status(void *aecInst, WebRtc_Word16 *status);
+
+/*
+ * Gets the current echo metrics for the session.
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void           *aecInst      Pointer to the AEC instance
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * AecMetrics     *metrics      Struct which will be filled out with the
+ *                              current echo metrics.
+ * WebRtc_Word32  return         0: OK
+ *                              -1: error
+ */
+WebRtc_Word32 WebRtcAec_GetMetrics(void *aecInst, AecMetrics *metrics);
+
+/*
+ * Gets the last error code.
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void           *aecInst      Pointer to the AEC instance
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word32  return        11000-11100: error code
+ */
+WebRtc_Word32 WebRtcAec_get_error_code(void *aecInst);
+
+/*
+ * Gets a version string.
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * char           *versionStr   Pointer to a string array
+ * WebRtc_Word16  len           The maximum length of the string
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word8   *versionStr   Pointer to a string array
+ * WebRtc_Word32  return         0: OK
+ *                              -1: error
+ */
+WebRtc_Word32 WebRtcAec_get_version(WebRtc_Word8 *versionStr, WebRtc_Word16 len);
+
+#ifdef __cplusplus
+}
+#endif
+#endif  /* WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_INTERFACE_ECHO_CANCELLATION_H_ */

src/modules/audio_processing/aec/main/matlab/fullaec.m

@@ -0,0 +1,953 @@
+% Partitioned block frequency domain adaptive filtering NLMS and 
+% standard time-domain sample-based NLMS 
+%fid=fopen('aecFar-samsung.pcm', 'rb'); % Load far end
+fid=fopen('aecFar.pcm', 'rb'); % Load far end
+%fid=fopen(farFile, 'rb'); % Load far end
+rrin=fread(fid,inf,'int16');
+fclose(fid); 
+%rrin=loadsl('data/far_me2.pcm'); % Load far end
+%fid=fopen('aecNear-samsung.pcm', 'rb'); % Load near end
+fid=fopen('aecNear.pcm', 'rb'); % Load near end
+%fid=fopen(nearFile, 'rb'); % Load near end
+ssin=fread(fid,inf,'int16');
+%ssin = [zeros(1024,1) ; ssin(1:end-1024)];
+
+fclose(fid);
+rand('state',13);
+fs=16000;
+mult=fs/8000;
+%rrin=rrin(fs*0+1:round(fs*120));
+%ssin=ssin(fs*0+1:round(fs*120));
+if fs == 8000
+	cohRange = 2:3;
+elseif fs==16000
+	cohRange = 2;
+end
+
+% Flags
+NLPon=1;  % NLP
+CNon=1; % Comfort noise
+PLTon=1;  % Plotting
+
+M = 16; % Number of partitions
+N = 64; % Partition length
+L = M*N; % Filter length 
+if fs == 8000
+    mufb = 0.6;
+else
+    mufb = 0.5;  
+end
+%mufb=1;  
+VADtd=48;
+alp = 0.1; % Power estimation factor alc = 0.1; % Coherence estimation factor
+beta = 0.9; % Plotting factor 
+%% Changed a little %%
+step = 0.3;%0.1875; % Downward step size 
+%%
+if fs == 8000
+    threshold=2e-6;  % DTrob threshold
+else
+    %threshold=0.7e-6;
+    threshold=1.5e-6; end
+
+if fs == 8000
+    echoBandRange = ceil(300*2/fs*N):floor(1800*2/fs*N);
+    %echoBandRange = ceil(1500*2/fs*N):floor(2500*2/fs*N);
+else
+    echoBandRange = ceil(300*2/fs*N):floor(1800*2/fs*N);
+    %echoBandRange = ceil(300*2/fs*N):floor(1800*2/fs*N);
+end
+%echoBandRange = ceil(1600*2/fs*N):floor(1900*2/fs*N);
+%echoBandRange = ceil(2000*2/fs*N):floor(4000*2/fs*N);
+suppState = 1;
+transCtr = 0;
+
+Nt=1;
+vt=1;
+
+ramp = 1.0003; % Upward ramp
+rampd = 0.999; % Downward ramp
+cvt = 20; % Subband VAD threshold;
+nnthres = 20; % Noise threshold 
+
+shh=logspace(-1.3,-2.2,N+1)';
+sh=[shh;flipud(shh(2:end-1))]; % Suppression profile
+
+len=length(ssin);
+w=zeros(L,1); % Sample-based TD NLMS 
+WFb=zeros(N+1,M); % Block-based FD NLMS
+WFbOld=zeros(N+1,M); % Block-based FD NLMS
+YFb=zeros(N+1,M);
+erfb=zeros(len,1);
+erfb3=zeros(len,1);
+
+ercn=zeros(len,1);
+zm=zeros(N,1);
+XFm=zeros(N+1,M);
+YFm=zeros(N+1,M);
+pn0=10*ones(N+1,1);
+pn=zeros(N+1,1);
+NN=len;
+Nb=floor(NN/N)-M;
+erifb=zeros(Nb+1,1)+0.1;
+erifb3=zeros(Nb+1,1)+0.1;
+ericn=zeros(Nb+1,1)+0.1;
+dri=zeros(Nb+1,1)+0.1;
+start=1;
+xo=zeros(N,1);
+do=xo;
+eo=xo;
+
+echoBands=zeros(Nb+1,1);
+cohxdAvg=zeros(Nb+1,1);
+cohxdSlow=zeros(Nb+1,N+1);
+cohedSlow=zeros(Nb+1,N+1);
+%overdriveM=zeros(Nb+1,N+1);
+cohxdFastAvg=zeros(Nb+1,1);
+cohxdAvgBad=zeros(Nb+1,1);
+cohedAvg=zeros(Nb+1,1);
+cohedFastAvg=zeros(Nb+1,1);
+hnledAvg=zeros(Nb+1,1);
+hnlxdAvg=zeros(Nb+1,1);
+ovrdV=zeros(Nb+1,1);
+dIdxV=zeros(Nb+1,1);
+SLxV=zeros(Nb+1,1);
+hnlSortQV=zeros(Nb+1,1);
+hnlPrefAvgV=zeros(Nb+1,1);
+mutInfAvg=zeros(Nb+1,1);
+%overdrive=zeros(Nb+1,1);
+hnled = zeros(N+1, 1);
+weight=zeros(N+1,1);
+hnlMax = zeros(N+1, 1);
+hnl = zeros(N+1, 1);
+overdrive = ones(1, N+1);
+xfwm=zeros(N+1,M);
+dfm=zeros(N+1,M);
+WFbD=ones(N+1,1);
+
+fbSupp = 0;
+hnlLocalMin = 1;
+cohxdLocalMin = 1;
+hnlLocalMinV=zeros(Nb+1,1);
+cohxdLocalMinV=zeros(Nb+1,1);
+hnlMinV=zeros(Nb+1,1);
+dkEnV=zeros(Nb+1,1);
+ekEnV=zeros(Nb+1,1);
+ovrd = 2;
+ovrdPos = floor((N+1)/4);
+ovrdSm = 2;
+hnlMin = 1;
+minCtr = 0;
+SeMin = 0;
+SdMin = 0;
+SeLocalAvg = 0;
+SeMinSm = 0;
+divergeFact = 1;
+dIdx = 1;
+hnlMinCtr = 0;
+hnlNewMin = 0;
+divergeState = 0;
+
+Sy=ones(N+1,1);
+Sym=1e7*ones(N+1,1);
+
+wins=[0;sqrt(hanning(2*N-1))];
+ubufn=zeros(2*N,1);
+ebuf=zeros(2*N,1);
+ebuf2=zeros(2*N,1);
+ebuf4=zeros(2*N,1);
+mbuf=zeros(2*N,1);
+
+cohedFast = zeros(N+1,1);
+cohxdFast = zeros(N+1,1);
+cohxd = zeros(N+1,1);
+Se = zeros(N+1,1);
+Sd = zeros(N+1,1);
+Sx = zeros(N+1,1);
+SxBad = zeros(N+1,1);
+Sed = zeros(N+1,1);
+Sxd = zeros(N+1,1);
+SxdBad = zeros(N+1,1);
+hnledp=[];
+
+cohxdMax = 0;
+
+%hh=waitbar(0,'Please wait...');
+progressbar(0);
+
+%spaces = ' ';
+%spaces = repmat(spaces, 50, 1);
+%spaces = ['[' ; spaces ; ']'];
+%fprintf(1, spaces);
+%fprintf(1, '\n');
+
+for kk=1:Nb
+    pos = N * (kk-1) + start;
+    
+    % FD block method
+    % ----------------------   Organize data
+    xk = rrin(pos:pos+N-1);
+    dk = ssin(pos:pos+N-1);
+    
+    xx = [xo;xk];
+    xo = xk;
+    tmp = fft(xx); 
+	XX = tmp(1:N+1);
+
+	dd = [do;dk];  % Overlap
+	do = dk;
+	tmp = fft(dd); % Frequency domain 
+	DD = tmp(1:N+1);
+    
+    % ------------------------  Power estimation
+    pn0 = (1 - alp) * pn0 + alp * real(XX.* conj(XX));
+    pn = pn0;
+    %pn = (1 - alp) * pn + alp * M * pn0;
+	if (CNon)
+		Yp =  real(conj(DD).*DD); % Instantaneous power      
+		Sy =  (1 - alp) * Sy + alp * Yp; % Averaged power    
+		
+		mm = min(Sy,Sym);  
+		diff = Sym - mm;
+		if (kk>50)
+			Sym = (mm + step*diff) * ramp; % Estimated background noise power   
+		end
+	end
+    
+    % ----------------------   Filtering   
+    XFm(:,1) = XX;
+    for mm=0:(M-1)
+        m=mm+1;  
+        YFb(:,m) = XFm(:,m) .* WFb(:,m);
+    end
+    yfk = sum(YFb,2);
+	tmp = [yfk ; flipud(conj(yfk(2:N)))];
+    ykt = real(ifft(tmp));
+    ykfb = ykt(end-N+1:end); 
+    
+    % ----------------------   Error estimation 
+    ekfb = dk - ykfb; 
+    %if sum(abs(ekfb)) < sum(abs(dk))
+        %ekfb = dk - ykfb; 
+    %    erfb(pos:pos+N-1) = ekfb; 
+    %else
+        %ekfb = dk;
+    %    erfb(pos:pos+N-1) = dk; 
+    %end
+	%(kk-1)*(N*2)+1
+    erfb(pos:pos+N-1) = ekfb; 
+    tmp = fft([zm;ekfb]);      % FD version for cancelling part (overlap-save)
+	Ek = tmp(1:N+1);
+
+    % ------------------------  Adaptation  
+	Ek2 = Ek ./(M*pn + 0.001); % Normalized error
+	%Ek2 = Ek ./(pn + 0.001); % Normalized error
+	%Ek2 = Ek ./(100*pn + 0.001); % Normalized error
+	
+	absEf = max(abs(Ek2), threshold);
+	absEf = ones(N+1,1)*threshold./absEf;
+	Ek2 = Ek2.*absEf;
+	
+	mEk = mufb.*Ek2;
+	PP = conj(XFm).*(ones(M,1) * mEk')'; 
+	tmp = [PP ; flipud(conj(PP(2:N,:)))];
+	IFPP = real(ifft(tmp));
+	PH = IFPP(1:N,:);
+	tmp = fft([PH;zeros(N,M)]);
+	FPH = tmp(1:N+1,:);
+	WFb = WFb + FPH;
+
+    if mod(kk, 10*mult) == 0
+        WFbEn = sum(real(WFb.*conj(WFb)));
+        %WFbEn = sum(abs(WFb));
+        [tmp, dIdx] = max(WFbEn);
+
+        WFbD = sum(abs(WFb(:, dIdx)),2);
+        %WFbD = WFbD / (mean(WFbD) + 1e-10); 
+        WFbD = min(max(WFbD, 0.5), 4);
+    end
+    dIdxV(kk) = dIdx;
+    
+	% NLP
+	if (NLPon)
+        
+        ee = [eo;ekfb]; 
+        eo = ekfb;
+		window = wins;
+        if fs == 8000
+            %gamma = 0.88;
+            gamma = 0.9;
+        else
+		    %gamma = 0.92;
+		    gamma = 0.93;
+        end
+		%gamma = 0.9;
+
+		tmp = fft(xx.*window);
+		xf = tmp(1:N+1);
+		tmp = fft(dd.*window);
+		df = tmp(1:N+1);
+		tmp = fft(ee.*window);
+		ef = tmp(1:N+1);
+
+        xfwm(:,1) = xf;
+        xf = xfwm(:,dIdx);
+        %fprintf(1,'%d: %f\n', kk, xf(4));
+        dfm(:,1) = df;
+        
+        SxOld = Sx;
+
+		Se = gamma*Se + (1-gamma)*real(ef.*conj(ef));
+		Sd = gamma*Sd + (1-gamma)*real(df.*conj(df));
+		Sx = gamma*Sx + (1 - gamma)*real(xf.*conj(xf));
+
+        %xRatio = real(xfwm(:,1).*conj(xfwm(:,1))) ./ ...
+        %    (real(xfwm(:,2).*conj(xfwm(:,2))) + 1e-10);
+        %xRatio = Sx ./ (SxOld + 1e-10);
+        %SLx = log(1/(N+1)*sum(xRatio)) - 1/(N+1)*sum(log(xRatio));
+        %SLxV(kk) = SLx;
+
+        %freqSm = 0.9;
+        %Sx = filter(freqSm, [1 -(1-freqSm)], Sx);
+        %Sx(end:1) = filter(freqSm, [1 -(1-freqSm)], Sx(end:1));
+        %Se = filter(freqSm, [1 -(1-freqSm)], Se);
+        %Se(end:1) = filter(freqSm, [1 -(1-freqSm)], Se(end:1));
+        %Sd = filter(freqSm, [1 -(1-freqSm)], Sd);
+        %Sd(end:1) = filter(freqSm, [1 -(1-freqSm)], Sd(end:1));
+
+		%SeFast = ef.*conj(ef);
+		%SdFast = df.*conj(df);
+        %SxFast = xf.*conj(xf);
+        %cohedFast = 0.9*cohedFast + 0.1*SeFast ./ (SdFast + 1e-10);
+        %cohedFast(find(cohedFast > 1)) = 1;
+        %cohedFast(find(cohedFast > 1)) = 1 ./ cohedFast(find(cohedFast>1));
+        %cohedFastAvg(kk) = mean(cohedFast(echoBandRange));
+        %cohedFastAvg(kk) = min(cohedFast);
+
+        %cohxdFast = 0.8*cohxdFast + 0.2*log(SdFast ./ (SxFast + 1e-10));
+        %cohxdFastAvg(kk) = mean(cohxdFast(echoBandRange));
+
+		% coherence
+        Sxd = gamma*Sxd + (1 - gamma)*xf.*conj(df);
+		Sed = gamma*Sed + (1-gamma)*ef.*conj(df);
+
+        %Sxd = filter(freqSm, [1 -(1-freqSm)], Sxd);
+        %Sxd(end:1) = filter(freqSm, [1 -(1-freqSm)], Sxd(end:1));
+        %Sed = filter(freqSm, [1 -(1-freqSm)], Sed);
+        %Sed(end:1) = filter(freqSm, [1 -(1-freqSm)], Sed(end:1));
+
+		cohed = real(Sed.*conj(Sed))./(Se.*Sd + 1e-10);
+        %cohedAvg(kk) = mean(cohed(echoBandRange));
+        %cohedAvg(kk) = cohed(6);
+        %cohedAvg(kk) = min(cohed);
+
+		cohxd = real(Sxd.*conj(Sxd))./(Sx.*Sd + 1e-10);
+        %freqSm = 0.5;
+        %cohxd(3:end) = filter(freqSm, [1 -(1-freqSm)], cohxd(3:end));
+        %cohxd(end:3) = filter(freqSm, [1 -(1-freqSm)], cohxd(end:3));
+        %cohxdAvg(kk) = mean(cohxd(echoBandRange));
+        %cohxdAvg(kk) = (cohxd(32));
+        %cohxdAvg(kk) = max(cohxd);
+
+        %xf = xfm(:,dIdx);
+		%SxBad = gamma*SxBad + (1 - gamma)*real(xf.*conj(xf));
+        %SxdBad = gamma*SxdBad + (1 - gamma)*xf.*conj(df);
+		%cohxdBad = real(SxdBad.*conj(SxdBad))./(SxBad.*Sd + 0.01);
+        %cohxdAvgBad(kk) = mean(cohxdBad);
+
+        %for j=1:N+1
+        %    mutInf(j) = 0.9*mutInf(j) + 0.1*information(abs(xfm(j,:)), abs(dfm(j,:)));
+        %end
+        %mutInfAvg(kk) = mean(mutInf);
+
+        %hnled = cohedFast;
+        %xIdx = find(cohxd > 1 - cohed);
+        %hnled(xIdx) = 1 - cohxd(xIdx);
+        %hnled = 1 - max(cohxd, 1-cohedFast);
+        hnled = min(1 - cohxd, cohed);
+        %hnled = 1 - cohxd;
+        %hnled = max(1 - (cohxd + (1-cohedFast)), 0);
+        %hnled = 1 - max(cohxd, 1-cohed);
+
+        if kk > 1
+            cohxdSlow(kk,:) = 0.99*cohxdSlow(kk-1,:) + 0.01*cohxd';
+            cohedSlow(kk,:) = 0.99*cohedSlow(kk-1,:) + 0.01*(1-cohed)';
+        end
+
+
+        if 0
+        %if kk > 50
+            %idx = find(hnled > 0.3);
+            hnlMax = hnlMax*0.9999;
+            %hnlMax(idx) = max(hnlMax(idx), hnled(idx));
+            hnlMax = max(hnlMax, hnled);
+            %overdrive(idx) = max(log(hnlMax(idx))/log(0.99), 1); 
+            avgHnl = mean(hnlMax(echoBandRange));
+            if avgHnl > 0.3
+                overdrive = max(log(avgHnl)/log(0.99), 1);
+            end
+            weight(4:end) = max(hnlMax) - hnlMax(4:end);
+        end
+        
+        
+
+        %[hg, gidx] = max(hnled); 
+        %fnrg = Sx(gidx) / (Sd(gidx) + 1e-10);
+        
+        %[tmp, bidx] = find((Sx / Sd + 1e-10) > fnrg);
+        %hnled(bidx) = hg;
+
+
+		%cohed1 = mean(cohed(cohRange)); % range depends on bandwidth 
+		%cohed1 = cohed1^2;
+        %echoBands(kk) = length(find(cohed(echoBandRange) < 0.25))/length(echoBandRange);
+
+        %if (fbSupp == 0)
+        %    if (echoBands(kk) > 0.8)
+        %        fbSupp = 1;
+        %    end
+        %else
+        %    if (echoBands(kk) < 0.6)
+        %        fbSupp = 0;
+        %    end
+        %end
+        %overdrive(kk) = 7.5*echoBands(kk) + 0.5; 
+        
+		% Factor by which to weight other bands
+		%if (cohed1 < 0.1)
+		%	w = 0.8 - cohed1*10*0.4;
+		%else
+		%	w = 0.4;
+		%end
+			
+		% Weight coherence subbands
+		%hnled = w*cohed1 + (1 - w)*cohed;
+		%hnled = (hnled).^2;
+		%cohed(floor(N/2):end) = cohed(floor(N/2):end).^2;
+        %if fbSupp == 1
+        %    cohed = zeros(size(cohed));
+        %end
+        %cohed = cohed.^overdrive(kk);
+
+        %hnled = gamma*hnled + (1 - gamma)*cohed;
+		% Additional hf suppression
+		%hnledp = [hnledp ; mean(hnled)];
+		%hnled(floor(N/2):end) = hnled(floor(N/2):end).^2;
+		%ef = ef.*((weight*(min(1 - hnled)).^2 + (1 - weight).*(1 - hnled)).^2);
+
+        cohedMean = mean(cohed(echoBandRange));
+        %aggrFact = 4*(1-mean(hnled(echoBandRange))) + 1;
+        %[hnlSort, hnlSortIdx] = sort(hnled(echoBandRange));
+        [hnlSort, hnlSortIdx] = sort(1-cohxd(echoBandRange));
+        [xSort, xSortIdx] = sort(Sx);
+        %aggrFact = (1-mean(hnled(echoBandRange)));
+        %hnlSortQ = hnlSort(qIdx);
+        hnlSortQ = mean(1 - cohxd(echoBandRange));
+        %hnlSortQ = mean(1 - cohxd);
+
+        [hnlSort2, hnlSortIdx2] = sort(hnled(echoBandRange));
+        %[hnlSort2, hnlSortIdx2] = sort(hnled);
+        hnlQuant = 0.75;
+        hnlQuantLow = 0.5;
+        qIdx = floor(hnlQuant*length(hnlSort2));
+        qIdxLow = floor(hnlQuantLow*length(hnlSort2));
+        hnlPrefAvg = hnlSort2(qIdx);
+        hnlPrefAvgLow = hnlSort2(qIdxLow);
+        %hnlPrefAvgLow = mean(hnled);
+        %hnlPrefAvg = max(hnlSort2);
+        %hnlPrefAvgLow = min(hnlSort2);
+
+        %hnlPref = hnled(echoBandRange);
+        %hnlPrefAvg = mean(hnlPref(xSortIdx((0.5*length(xSortIdx)):end)));
+
+        %hnlPrefAvg = min(hnlPrefAvg, hnlSortQ);
+
+        %hnlSortQIdx = hnlSortIdx(qIdx);
+        %SeQ = Se(qIdx + echoBandRange(1) - 1); 
+        %SdQ = Sd(qIdx + echoBandRange(1) - 1); 
+        %SeQ = Se(qIdxLow + echoBandRange(1) - 1); 
+        %SdQ = Sd(qIdxLow + echoBandRange(1) - 1); 
+        %propLow = length(find(hnlSort < 0.1))/length(hnlSort);
+        %aggrFact = min((1 - hnlSortQ)/2, 0.5);
+        %aggrTerm = 1/aggrFact;
+
+        %hnlg = mean(hnled(echoBandRange));
+        %hnlg = hnlSortQ;
+        %if suppState == 0
+        %    if hnlg < 0.05
+        %        suppState = 2;
+        %        transCtr = 0;
+        %    elseif hnlg < 0.75
+        %        suppState = 1;
+        %        transCtr = 0;
+        %    end
+        %elseif suppState == 1
+        %    if hnlg > 0.8
+        %        suppState = 0;
+        %        transCtr = 0;
+        %    elseif hnlg < 0.05
+        %        suppState = 2;
+        %        transCtr = 0;
+        %    end
+        %else
+        %    if hnlg > 0.8
+        %        suppState = 0;
+        %        transCtr = 0;
+        %    elseif hnlg > 0.25
+        %        suppState = 1;
+        %        transCtr = 0;
+        %    end
+        %end
+        %if kk > 50
+
+            if cohedMean > 0.98 & hnlSortQ > 0.9
+                %if suppState == 1
+                %    hnled = 0.5*hnled + 0.5*cohed;
+                %    %hnlSortQ = 0.5*hnlSortQ + 0.5*cohedMean;
+                %    hnlPrefAvg = 0.5*hnlPrefAvg + 0.5*cohedMean;
+                %else
+                %    hnled = cohed;
+                %    %hnlSortQ = cohedMean;
+                %    hnlPrefAvg = cohedMean;
+                %end
+                suppState = 0;
+            elseif cohedMean < 0.95 | hnlSortQ < 0.8
+                %if suppState == 0
+                %    hnled = 0.5*hnled + 0.5*cohed;
+                %    %hnlSortQ = 0.5*hnlSortQ + 0.5*cohedMean;
+                %    hnlPrefAvg = 0.5*hnlPrefAvg + 0.5*cohedMean;
+                %end
+                suppState = 1;
+            end
+
+            if hnlSortQ < cohxdLocalMin & hnlSortQ < 0.75
+                cohxdLocalMin = hnlSortQ;
+            end
+
+            if cohxdLocalMin == 1
+                ovrd = 3;
+                hnled = 1-cohxd;
+                hnlPrefAvg = hnlSortQ;
+                hnlPrefAvgLow = hnlSortQ;
+            end
+
+            if suppState == 0
+                hnled = cohed;
+                hnlPrefAvg = cohedMean;
+                hnlPrefAvgLow = cohedMean;
+            end
+
+            %if hnlPrefAvg < hnlLocalMin & hnlPrefAvg < 0.6
+            if hnlPrefAvgLow < hnlLocalMin & hnlPrefAvgLow < 0.6
+                %hnlLocalMin = hnlPrefAvg;
+                %hnlMin = hnlPrefAvg;
+                hnlLocalMin = hnlPrefAvgLow;
+                hnlMin = hnlPrefAvgLow;
+                hnlNewMin = 1;
+                hnlMinCtr = 0;
+                %if hnlMinCtr == 0
+                %    hnlMinCtr = hnlMinCtr + 1;
+                %else
+                %    hnlMinCtr = 0;
+                %    hnlMin = hnlLocalMin;
+                    %SeLocalMin = SeQ;
+                    %SdLocalMin = SdQ;
+                    %SeLocalAvg = 0;
+                    %minCtr = 0;
+                %    ovrd = max(log(0.0001)/log(hnlMin), 2);
+                    %divergeFact = hnlLocalMin;
+            end
+
+            if hnlNewMin == 1
+                hnlMinCtr = hnlMinCtr + 1;
+            end
+            if hnlMinCtr == 2 
+                hnlNewMin = 0;
+                hnlMinCtr = 0;
+                %ovrd = max(log(0.0001)/log(hnlMin), 2);
+                ovrd = max(log(0.00001)/(log(hnlMin + 1e-10) + 1e-10), 3);
+                %ovrd = max(log(0.00000001)/(log(hnlMin + 1e-10) + 1e-10), 5);
+                %ovrd = max(log(0.0001)/log(hnlPrefAvg), 2);
+                %ovrd = max(log(0.001)/log(hnlMin), 2);
+            end
+            hnlLocalMin = min(hnlLocalMin + 0.0008/mult, 1);
+            cohxdLocalMin = min(cohxdLocalMin + 0.0004/mult, 1);
+            %divergeFact = hnlSortQ;
+
+
+            %if minCtr > 0 & hnlLocalMin < 1
+            %    hnlMin = hnlLocalMin;
+            %    %SeMin = 0.9*SeMin + 0.1*sqrt(SeLocalMin);
+            %    SdMin = sqrt(SdLocalMin);
+            %    %SeMin = sqrt(SeLocalMin)*hnlSortQ;
+            %    SeMin = sqrt(SeLocalMin);
+            %    %ovrd = log(100/SeMin)/log(hnlSortQ);
+            %    %ovrd = log(100/SeMin)/log(hnlSortQ);
+            %    ovrd = log(0.01)/log(hnlMin);
+            %    ovrd = max(ovrd, 2);
+            %    ovrdPos = hnlSortQIdx;
+            %    %ovrd = max(ovrd, 1);
+            %    %SeMin = sqrt(SeLocalAvg/5);
+            %    minCtr = 0;
+            %else
+            %    %SeLocalMin = 0.9*SeLocalMin +0.1*SeQ;
+            %    SeLocalAvg = SeLocalAvg + SeQ;
+            %    minCtr = minCtr + 1;
+            %end
+
+            if ovrd < ovrdSm
+                ovrdSm = 0.99*ovrdSm + 0.01*ovrd;
+            else
+                ovrdSm = 0.9*ovrdSm + 0.1*ovrd;
+            end
+        %end
+
+        %ekEn = sum(real(ekfb.^2));
+        %dkEn = sum(real(dk.^2));
+        ekEn = sum(Se);
+        dkEn = sum(Sd);
+
+        if divergeState == 0
+            if ekEn > dkEn
+                ef = df;
+                divergeState = 1;
+                %hnlPrefAvg = hnlSortQ;
+                %hnled = (1 - cohxd);
+            end
+        else
+            %if ekEn*1.1 < dkEn  
+            %if ekEn*1.26 < dkEn  
+            if ekEn*1.05 < dkEn  
+                divergeState = 0;
+            else
+                ef = df;
+            end
+        end
+
+        if ekEn > dkEn*19.95
+            WFb=zeros(N+1,M); % Block-based FD NLMS
+        end
+
+        ekEnV(kk) = ekEn;
+        dkEnV(kk) = dkEn;
+
+        hnlLocalMinV(kk) = hnlLocalMin;
+        cohxdLocalMinV(kk) = cohxdLocalMin;
+        hnlMinV(kk) = hnlMin;
+        %cohxdMaxLocal = max(cohxdSlow(kk,:));
+        %if kk > 50
+        %cohxdMaxLocal = 1-hnlSortQ;
+        %if cohxdMaxLocal > 0.5
+        %    %if cohxdMaxLocal > cohxdMax 
+        %        odScale = max(log(cohxdMaxLocal)/log(0.95), 1);
+        %        %overdrive(7:end) = max(log(cohxdSlow(kk,7:end))/log(0.9), 1);
+        %        cohxdMax = cohxdMaxLocal;
+        %    end
+        %end
+        %end
+        %cohxdMax = cohxdMax*0.999;
+
+        %overdriveM(kk,:) = max(overdrive, 1);
+        %aggrFact = 0.25;
+        aggrFact = 0.3;
+        %aggrFact = 0.5*propLow;
+        %if fs == 8000
+        %    wCurve = [0 ; 0 ; aggrFact*sqrt(linspace(0,1,N-1))' + 0.1];
+        %else
+        %    wCurve = [0; 0; 0; aggrFact*sqrt(linspace(0,1,N-2))' + 0.1];
+        %end
+        wCurve = [0; aggrFact*sqrt(linspace(0,1,N))' + 0.1];
+        % For sync with C
+        %if fs == 8000
+        %    wCurve = wCurve(2:end);
+        %else
+        %    wCurve = wCurve(1:end-1);
+        %end
+        %weight = aggrFact*(sqrt(linspace(0,1,N+1)'));
+        %weight = aggrFact*wCurve;
+        weight = wCurve;
+        %weight = aggrFact*ones(N+1,1);
+        %weight = zeros(N+1,1);
+        %hnled = weight.*min(hnled) + (1 - weight).*hnled;
+        %hnled = weight.*min(mean(hnled(echoBandRange)), hnled) + (1 - weight).*hnled;
+        %hnled = weight.*min(hnlSortQ, hnled) + (1 - weight).*hnled;
+
+        %hnlSortQV(kk) = mean(hnled);
+        %hnlPrefAvgV(kk) = mean(hnled(echoBandRange));
+
+        hnled = weight.*min(hnlPrefAvg, hnled) + (1 - weight).*hnled;
+
+        %od = aggrFact*(sqrt(linspace(0,1,N+1)') + aggrTerm);
+        %od = 4*(sqrt(linspace(0,1,N+1)') + 1/4);
+
+        %ovrdFact = (ovrdSm - 1) / sqrt(ovrdPos/(N+1));
+        %ovrdFact = ovrdSm / sqrt(echoBandRange(floor(length(echoBandRange)/2))/(N+1));
+        %od = ovrdFact*sqrt(linspace(0,1,N+1))' + 1;
+        %od = ovrdSm*ones(N+1,1).*abs(WFb(:,dIdx))/(max(abs(WFb(:,dIdx)))+1e-10);
+
+        %od = ovrdSm*ones(N+1,1);
+        %od = ovrdSm*WFbD.*(sqrt(linspace(0,1,N+1))' + 1);
+
+        od = ovrdSm*(sqrt(linspace(0,1,N+1))' + 1);
+        %od = 4*(sqrt(linspace(0,1,N+1))' + 1);
+
+        %od = 2*ones(N+1,1);
+        %od = 2*ones(N+1,1);
+        %sshift = ((1-hnled)*2-1).^3+1; 
+        sshift = ones(N+1,1);
+
+        hnled = hnled.^(od.*sshift);
+
+        %if hnlg > 0.75
+            %if (suppState ~= 0)
+            %    transCtr = 0;
+            %end
+        %    suppState = 0;
+        %elseif hnlg < 0.6 & hnlg > 0.2
+        %    suppState = 1;
+        %elseif hnlg < 0.1
+            %hnled = zeros(N+1, 1);
+            %if (suppState ~= 2)
+            %    transCtr = 0;
+            %end
+        %    suppState = 2;
+        %else
+        %    if (suppState ~= 2)
+        %        transCtr = 0;
+        %    end
+        %    suppState = 2;
+        %end
+        %if suppState == 0
+        %    hnled = ones(N+1, 1);
+        %elseif suppState == 2
+        %    hnled = zeros(N+1, 1);
+        %end
+        %hnled(find(hnled < 0.1)) = 0;
+        %hnled = hnled.^2;
+        %if transCtr < 5
+            %hnl = 0.75*hnl + 0.25*hnled;
+        %    transCtr = transCtr + 1;
+        %else
+            hnl = hnled;
+        %end
+        %hnled(find(hnled < 0.05)) = 0;
+        ef = ef.*(hnl);
+
+        %ef = ef.*(min(1 - cohxd, cohed).^2);
+        %ef = ef.*((1-cohxd).^2);
+        
+        ovrdV(kk) = ovrdSm;
+        %ovrdV(kk) = dIdx;
+        %ovrdV(kk) = divergeFact;
+        %hnledAvg(kk) = 1-mean(1-cohedFast(echoBandRange));
+        hnledAvg(kk) = 1-mean(1-cohed(echoBandRange));
+        hnlxdAvg(kk) = 1-mean(cohxd(echoBandRange));
+        %hnlxdAvg(kk) = cohxd(5);
+        %hnlSortQV(kk) = mean(hnled);
+        hnlSortQV(kk) = hnlPrefAvgLow;
+        hnlPrefAvgV(kk) = hnlPrefAvg;
+        %hnlAvg(kk) = propLow;
+        %ef(N/2:end) = 0;
+        %ner = (sum(Sd) ./ (sum(Se.*(hnl.^2)) + 1e-10));
+
+		% Comfort noise
+		if (CNon)
+			snn=sqrt(Sym);
+			snn(1)=0; % Reject LF noise
+			Un=snn.*exp(j*2*pi.*[0;rand(N-1,1);0]);
+
+			% Weight comfort noise by suppression
+			Un = sqrt(1-hnled.^2).*Un;
+			Fmix = ef + Un;
+		else
+			Fmix = ef;
+		end
+
+		% Overlap and add in time domain for smoothness 
+		tmp = [Fmix ; flipud(conj(Fmix(2:N)))];
+		mixw = wins.*real(ifft(tmp));
+		mola  = mbuf(end-N+1:end) + mixw(1:N);
+		mbuf = mixw;
+		ercn(pos:pos+N-1) = mola; 
+	end % NLPon
+
+    % Filter update
+	%Ek2 = Ek ./(12*pn + 0.001); % Normalized error
+    %Ek2 = Ek2 * divergeFact;
+	%Ek2 = Ek ./(pn + 0.001); % Normalized error
+	%Ek2 = Ek ./(100*pn + 0.001); % Normalized error
+
+    %divergeIdx = find(abs(Ek) > abs(DD));
+    %divergeIdx = find(Se > Sd);
+    %threshMod = threshold*ones(N+1,1); 
+    %if length(divergeIdx) > 0
+    %if sum(abs(Ek)) > sum(abs(DD))
+        %WFb(divergeIdx,:) = WFb(divergeIdx,:) .* repmat(sqrt(Sd(divergeIdx)./(Se(divergeIdx)+1e-10))),1,M);
+        %Ek2(divergeIdx) = Ek2(divergeIdx) .* sqrt(Sd(divergeIdx)./(Se(divergeIdx)+1e-10));
+        %Ek2(divergeIdx) = Ek2(divergeIdx) .* abs(DD(divergeIdx))./(abs(Ek(divergeIdx))+1e-10);
+        %WFb(divergeIdx,:) = WFbOld(divergeIdx,:);
+        %WFb = WFbOld;
+        %threshMod(divergeIdx) = threshMod(divergeIdx) .* abs(DD(divergeIdx))./(abs(Ek(divergeIdx))+1e-10);
+    %    threshMod(divergeIdx) = threshMod(divergeIdx) .* sqrt(Sd(divergeIdx)./(Se(divergeIdx)+1e-10));
+    %end
+	
+	%absEf = max(abs(Ek2), threshold);
+	%absEf = ones(N+1,1)*threshold./absEf;
+	%absEf = max(abs(Ek2), threshMod);
+	%absEf = threshMod./absEf;
+	%Ek2 = Ek2.*absEf;
+
+    %if sum(Se) <= sum(Sd)
+
+    %    mEk = mufb.*Ek2;
+    %    PP = conj(XFm).*(ones(M,1) * mEk')'; 
+    %    tmp = [PP ; flipud(conj(PP(2:N,:)))];
+    %    IFPP = real(ifft(tmp));
+    %    PH = IFPP(1:N,:);
+    %    tmp = fft([PH;zeros(N,M)]);
+    %    FPH = tmp(1:N+1,:);
+    %    %WFbOld = WFb;
+    %    WFb = WFb + FPH;
+
+    %else
+    %    WF = WFbOld;
+    %end
+
+	% Shift old FFTs
+	%for m=M:-1:2
+	%	XFm(:,m) = XFm(:,m-1);    
+	%	YFm(:,m) = YFm(:,m-1);    
+	%end
+    XFm(:,2:end) = XFm(:,1:end-1);
+    YFm(:,2:end) = YFm(:,1:end-1);
+    xfwm(:,2:end) = xfwm(:,1:end-1);
+    dfm(:,2:end) = dfm(:,1:end-1);
+
+	%if mod(kk, floor(Nb/50)) == 0
+    %    fprintf(1, '.');
+	%end
+
+	if mod(kk, floor(Nb/100)) == 0
+	%if mod(kk, floor(Nb/500)) == 0
+        progressbar(kk/Nb); 
+        %figure(5)
+        %plot(abs(WFb));
+        %legend('1','2','3','4','5','6','7','8','9','10','11','12');
+        %title(kk*N/fs);
+        %figure(6)
+        %plot(WFbD);
+        %figure(6)
+        %plot(threshMod)
+        %if length(divergeIdx) > 0
+        %    plot(abs(DD))
+        %    hold on
+        %    plot(abs(Ek), 'r')
+        %    hold off
+            %plot(min(sqrt(Sd./(Se+1e-10)),1))
+            %axis([0 N 0 1]);
+        %end
+        %figure(6)
+        %plot(cohedFast);
+        %axis([1 N+1 0 1]);
+        %plot(WFbEn);
+
+        %figure(7)
+        %plot(weight);
+        %plot([cohxd 1-cohed]);
+        %plot([cohxd 1-cohed 1-cohedFast hnled]);
+        %plot([cohxd cohxdFast/max(cohxdFast)]);
+        %legend('cohxd', '1-cohed', '1-cohedFast');
+        %axis([1 65 0 1]);
+        %pause(0.5);
+        %overdrive
+    end
+end
+progressbar(1);
+
+%figure(2);
+%plot([feat(:,1) feat(:,2)+1 feat(:,3)+2 mfeat+3]);
+%plot([feat(:,1) mfeat+1]);
+
+%figure(3);
+%plot(10*log10([dri erifb erifb3 ericn]));
+%legend('Near-end','Error','Post NLP','Final',4);
+% Compensate for delay
+%ercn=[ercn(N+1:end);zeros(N,1)];
+%ercn_=[ercn_(N+1:end);zeros(N,1)];
+
+%figure(11);
+%plot(cohxdSlow);
+
+%figure(12);
+%surf(cohxdSlow);
+%shading interp;
+
+%figure(13);
+%plot(overdriveM);
+
+%figure(14);
+%surf(overdriveM);
+%shading interp;
+
+figure(10);
+t = (0:Nb)*N/fs;
+rrinSubSamp = rrin(N*(1:(Nb+1)));
+plot(t, rrinSubSamp/max(abs(rrinSubSamp)),'b');
+hold on
+plot(t, hnledAvg, 'r');
+plot(t, hnlxdAvg, 'g');
+plot(t, hnlSortQV, 'y');
+plot(t, hnlLocalMinV, 'k');
+plot(t, cohxdLocalMinV, 'c');
+plot(t, hnlPrefAvgV, 'm');
+%plot(t, cohxdAvg, 'r');
+%plot(cohxdFastAvg, 'r');
+%plot(cohxdAvgBad, 'k');
+%plot(t, cohedAvg, 'k');
+%plot(t, 1-cohedFastAvg, 'k');
+%plot(ssin(N*(1:floor(length(ssin)/N)))/max(abs(ssin)));
+%plot(echoBands,'r');
+%plot(overdrive, 'g');
+%plot(erfb(N*(1:floor(length(erfb)/N)))/max(abs(erfb)));
+hold off
+tightx;
+
+figure(11)
+plot(t, ovrdV);
+tightx;
+%plot(mfeat,'r');
+%plot(1-cohxyp_,'r');
+%plot(Hnlxydp,'y');
+%plot(hnledp,'k');
+%plot(Hnlxydp, 'c');
+%plot(ccohpd_,'k');
+%plot(supplot_, 'g');
+%plot(ones(length(mfeat),1)*rr1_, 'k');
+%plot(ones(length(mfeat),1)*rr2_, 'k');
+%plot(N*(1:length(feat)), feat);
+%plot(Sep_,'r');
+%axis([1 floor(length(erfb)/N) -1 1])
+%hold off
+%plot(10*log10([Se_, Sx_, Seu_, real(sf_.*conj(sf_))]));
+%legend('Se','Sx','Seu','S');
+%figure(5)
+%plot([ercn ercn_]);
+
+figure(12)
+plot(t, dIdxV);
+%plot(t, SLxV);
+tightx;
+
+%figure(13)
+%plot(t, [ekEnV dkEnV]);
+%plot(t, dkEnV./(ekEnV+1e-10));
+%tightx;
+
+%close(hh);
+%spclab(fs,ssin,erfb,ercn,'outxd.pcm');
+%spclab(fs,rrin,ssin,erfb,1.78*ercn,'vqeOut-1.pcm');
+%spclab(fs,erfb,'aecOutLp.pcm');
+%spclab(fs,rrin,ssin,erfb,1.78*ercn,'aecOut25.pcm','vqeOut-1.pcm');
+%spclab(fs,rrin,ssin,erfb,ercn,'aecOut-mba.pcm');
+%spclab(fs,rrin,ssin,erfb,ercn,'aecOut.pcm');
+%spclab(fs, ssin, erfb, ercn, 'out0.pcm');

src/modules/audio_processing/aec/main/source/aec.gypi

@@ -0,0 +1,46 @@
+# Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+#
+# Use of this source code is governed by a BSD-style license
+# that can be found in the LICENSE file in the root of the source
+# tree. An additional intellectual property rights grant can be found
+# in the file PATENTS.  All contributing project authors may
+# be found in the AUTHORS file in the root of the source tree.
+
+{
+  'targets': [
+    {
+      'target_name': 'aec',
+      'type': '<(library)',
+      'dependencies': [
+        '<(webrtc_root)/common_audio/common_audio.gyp:spl',
+        'apm_util'
+      ],
+      'include_dirs': [
+        '../interface',
+      ],
+      'direct_dependent_settings': {
+        'include_dirs': [
+          '../interface',
+        ],
+      },
+      'sources': [
+        '../interface/echo_cancellation.h',
+        'echo_cancellation.c',
+        'aec_core.h',
+        'aec_core.c',
+        'aec_core_sse2.c',
+        'aec_rdft.h',
+        'aec_rdft.c',
+        'aec_rdft_sse2.c',
+        'resampler.h',
+        'resampler.c',
+      ],
+    },
+  ],
+}
+
+# Local Variables:
+# tab-width:2
+# indent-tabs-mode:nil
+# End:
+# vim: set expandtab tabstop=2 shiftwidth=2:

src/modules/audio_processing/aec/main/source/aec_core.h

@@ -0,0 +1,176 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+/*
+ * Specifies the interface for the AEC core.
+ */
+
+#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_AEC_CORE_H_
+#define WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_AEC_CORE_H_
+
+#include <stdio.h>
+
+#include "signal_processing_library.h"
+#include "typedefs.h"
+
+//#define UNCONSTR // time-unconstrained filter
+//#define AEC_DEBUG // for recording files
+
+#define FRAME_LEN 80
+#define PART_LEN 64 // Length of partition
+#define PART_LEN1 (PART_LEN + 1) // Unique fft coefficients
+#define PART_LEN2 (PART_LEN * 2) // Length of partition * 2
+#define NR_PART 12 // Number of partitions
+#define FILT_LEN (PART_LEN * NR_PART) // Filter length
+#define FILT_LEN2 (FILT_LEN * 2) // Double filter length
+#define FAR_BUF_LEN (FILT_LEN2 * 2)
+#define PREF_BAND_SIZE 24
+
+#define BLOCKL_MAX FRAME_LEN
+
+typedef float complex_t[2];
+// For performance reasons, some arrays of complex numbers are replaced by twice
+// as long arrays of float, all the real parts followed by all the imaginary
+// ones (complex_t[SIZE] -> float[2][SIZE]). This allows SIMD optimizations and
+// is better than two arrays (one for the real parts and one for the imaginary
+// parts) as this other way would require two pointers instead of one and cause
+// extra register spilling. This also allows the offsets to be calculated at
+// compile time.
+
+// Metrics
+enum {offsetLevel = -100};
+
+typedef struct {
+    float sfrsum;
+    int sfrcounter;
+    float framelevel;
+    float frsum;
+    int frcounter;
+    float minlevel;
+    float averagelevel;
+} power_level_t;
+
+typedef struct {
+    float instant;
+    float average;
+    float min;
+    float max;
+    float sum;
+    float hisum;
+    float himean;
+    int counter;
+    int hicounter;
+} stats_t;
+
+typedef struct {
+    int farBufWritePos, farBufReadPos;
+
+    int knownDelay;
+    int inSamples, outSamples;
+    int delayEstCtr;
+
+    void *farFrBuf, *nearFrBuf, *outFrBuf;
+
+    void *nearFrBufH;
+    void *outFrBufH;
+
+    float xBuf[PART_LEN2]; // farend
+    float dBuf[PART_LEN2]; // nearend
+    float eBuf[PART_LEN2]; // error
+
+    float dBufH[PART_LEN2]; // nearend
+
+    float xPow[PART_LEN1];
+    float dPow[PART_LEN1];
+    float dMinPow[PART_LEN1];
+    float dInitMinPow[PART_LEN1];
+    float *noisePow;
+
+    float xfBuf[2][NR_PART * PART_LEN1]; // farend fft buffer
+    float wfBuf[2][NR_PART * PART_LEN1]; // filter fft
+    complex_t sde[PART_LEN1]; // cross-psd of nearend and error
+    complex_t sxd[PART_LEN1]; // cross-psd of farend and nearend
+    complex_t xfwBuf[NR_PART * PART_LEN1]; // farend windowed fft buffer
+
+    float sx[PART_LEN1], sd[PART_LEN1], se[PART_LEN1]; // far, near and error psd
+    float hNs[PART_LEN1];
+    float hNlFbMin, hNlFbLocalMin;
+    float hNlXdAvgMin;
+    int hNlNewMin, hNlMinCtr;
+    float overDrive, overDriveSm;
+    float targetSupp, minOverDrive;
+    float outBuf[PART_LEN];
+    int delayIdx;
+
+    short stNearState, echoState;
+    short divergeState;
+
+    int xfBufBlockPos;
+
+    short farBuf[FILT_LEN2 * 2];
+
+    short mult; // sampling frequency multiple
+    int sampFreq;
+    WebRtc_UWord32 seed;
+
+    float mu; // stepsize
+    float errThresh; // error threshold
+
+    int noiseEstCtr;
+
+    power_level_t farlevel;
+    power_level_t nearlevel;
+    power_level_t linoutlevel;
+    power_level_t nlpoutlevel;
+
+    int metricsMode;
+    int stateCounter;
+    stats_t erl;
+    stats_t erle;
+    stats_t aNlp;
+    stats_t rerl;
+
+    // Quantities to control H band scaling for SWB input
+    int freq_avg_ic;         //initial bin for averaging nlp gain
+    int flag_Hband_cn;      //for comfort noise
+    float cn_scale_Hband;   //scale for comfort noise in H band
+
+#ifdef AEC_DEBUG
+    FILE *farFile;
+    FILE *nearFile;
+    FILE *outFile;
+    FILE *outLpFile;
+#endif
+} aec_t;
+
+typedef void (*WebRtcAec_FilterFar_t)(aec_t *aec, float yf[2][PART_LEN1]);
+extern WebRtcAec_FilterFar_t WebRtcAec_FilterFar;
+typedef void (*WebRtcAec_ScaleErrorSignal_t)(aec_t *aec, float ef[2][PART_LEN1]);
+extern WebRtcAec_ScaleErrorSignal_t WebRtcAec_ScaleErrorSignal;
+typedef void (*WebRtcAec_FilterAdaptation_t)
+  (aec_t *aec, float *fft, float ef[2][PART_LEN1]);
+extern WebRtcAec_FilterAdaptation_t WebRtcAec_FilterAdaptation;
+typedef void (*WebRtcAec_OverdriveAndSuppress_t)
+  (aec_t *aec, float hNl[PART_LEN1], const float hNlFb, float efw[2][PART_LEN1]);
+extern WebRtcAec_OverdriveAndSuppress_t WebRtcAec_OverdriveAndSuppress;
+
+int WebRtcAec_CreateAec(aec_t **aec);
+int WebRtcAec_FreeAec(aec_t *aec);
+int WebRtcAec_InitAec(aec_t *aec, int sampFreq);
+void WebRtcAec_InitAec_SSE2(void);
+
+void WebRtcAec_InitMetrics(aec_t *aec);
+void WebRtcAec_ProcessFrame(aec_t *aec, const short *farend,
+                       const short *nearend, const short *nearendH,
+                       short *out, short *outH,
+                       int knownDelay);
+
+#endif // WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_AEC_CORE_H_
+

src/modules/audio_processing/aec/main/source/aec_core_sse2.c

@@ -0,0 +1,429 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+/*
+ * The core AEC algorithm, SSE2 version of speed-critical functions.
+ */
+
+#include "typedefs.h"
+
+#if defined(WEBRTC_USE_SSE2)
+#include <emmintrin.h>
+#include <math.h>
+
+#include "aec_core.h"
+#include "aec_rdft.h"
+
+__inline static float MulRe(float aRe, float aIm, float bRe, float bIm)
+{
+  return aRe * bRe - aIm * bIm;
+}
+
+__inline static float MulIm(float aRe, float aIm, float bRe, float bIm)
+{
+  return aRe * bIm + aIm * bRe;
+}
+
+static void FilterFarSSE2(aec_t *aec, float yf[2][PART_LEN1])
+{
+  int i;
+  for (i = 0; i < NR_PART; i++) {
+    int j;
+    int xPos = (i + aec->xfBufBlockPos) * PART_LEN1;
+    int pos = i * PART_LEN1;
+    // Check for wrap
+    if (i + aec->xfBufBlockPos >= NR_PART) {
+      xPos -= NR_PART*(PART_LEN1);
+    }
+
+    // vectorized code (four at once)
+    for (j = 0; j + 3 < PART_LEN1; j += 4) {
+      const __m128 xfBuf_re = _mm_loadu_ps(&aec->xfBuf[0][xPos + j]);
+      const __m128 xfBuf_im = _mm_loadu_ps(&aec->xfBuf[1][xPos + j]);
+      const __m128 wfBuf_re = _mm_loadu_ps(&aec->wfBuf[0][pos + j]);
+      const __m128 wfBuf_im = _mm_loadu_ps(&aec->wfBuf[1][pos + j]);
+      const __m128 yf_re = _mm_loadu_ps(&yf[0][j]);
+      const __m128 yf_im = _mm_loadu_ps(&yf[1][j]);
+      const __m128 a = _mm_mul_ps(xfBuf_re, wfBuf_re);
+      const __m128 b = _mm_mul_ps(xfBuf_im, wfBuf_im);
+      const __m128 c = _mm_mul_ps(xfBuf_re, wfBuf_im);
+      const __m128 d = _mm_mul_ps(xfBuf_im, wfBuf_re);
+      const __m128 e = _mm_sub_ps(a, b);
+      const __m128 f = _mm_add_ps(c, d);
+      const __m128 g = _mm_add_ps(yf_re, e);
+      const __m128 h = _mm_add_ps(yf_im, f);
+      _mm_storeu_ps(&yf[0][j], g);
+      _mm_storeu_ps(&yf[1][j], h);
+    }
+    // scalar code for the remaining items.
+    for (; j < PART_LEN1; j++) {
+      yf[0][j] += MulRe(aec->xfBuf[0][xPos + j], aec->xfBuf[1][xPos + j],
+                        aec->wfBuf[0][ pos + j], aec->wfBuf[1][ pos + j]);
+      yf[1][j] += MulIm(aec->xfBuf[0][xPos + j], aec->xfBuf[1][xPos + j],
+                        aec->wfBuf[0][ pos + j], aec->wfBuf[1][ pos + j]);
+    }
+  }
+}
+
+static void ScaleErrorSignalSSE2(aec_t *aec, float ef[2][PART_LEN1])
+{
+  const __m128 k1e_10f = _mm_set1_ps(1e-10f);
+  const __m128 kThresh = _mm_set1_ps(aec->errThresh);
+  const __m128 kMu = _mm_set1_ps(aec->mu);
+
+  int i;
+  // vectorized code (four at once)
+  for (i = 0; i + 3 < PART_LEN1; i += 4) {
+    const __m128 xPow = _mm_loadu_ps(&aec->xPow[i]);
+    const __m128 ef_re_base = _mm_loadu_ps(&ef[0][i]);
+    const __m128 ef_im_base = _mm_loadu_ps(&ef[1][i]);
+
+    const __m128 xPowPlus = _mm_add_ps(xPow, k1e_10f);
+    __m128 ef_re = _mm_div_ps(ef_re_base, xPowPlus);
+    __m128 ef_im = _mm_div_ps(ef_im_base, xPowPlus);
+    const __m128 ef_re2 = _mm_mul_ps(ef_re, ef_re);
+    const __m128 ef_im2 = _mm_mul_ps(ef_im, ef_im);
+    const __m128 ef_sum2 = _mm_add_ps(ef_re2, ef_im2);
+    const __m128 absEf = _mm_sqrt_ps(ef_sum2);
+    const __m128 bigger = _mm_cmpgt_ps(absEf, kThresh);
+    __m128 absEfPlus = _mm_add_ps(absEf, k1e_10f);
+    const __m128 absEfInv = _mm_div_ps(kThresh, absEfPlus);
+    __m128 ef_re_if = _mm_mul_ps(ef_re, absEfInv);
+    __m128 ef_im_if = _mm_mul_ps(ef_im, absEfInv);
+    ef_re_if = _mm_and_ps(bigger, ef_re_if);
+    ef_im_if = _mm_and_ps(bigger, ef_im_if);
+    ef_re = _mm_andnot_ps(bigger, ef_re);
+    ef_im = _mm_andnot_ps(bigger, ef_im);
+    ef_re = _mm_or_ps(ef_re, ef_re_if);
+    ef_im = _mm_or_ps(ef_im, ef_im_if);
+    ef_re = _mm_mul_ps(ef_re, kMu);
+    ef_im = _mm_mul_ps(ef_im, kMu);
+
+    _mm_storeu_ps(&ef[0][i], ef_re);
+    _mm_storeu_ps(&ef[1][i], ef_im);
+  }
+  // scalar code for the remaining items.
+  for (; i < (PART_LEN1); i++) {
+    float absEf;
+    ef[0][i] /= (aec->xPow[i] + 1e-10f);
+    ef[1][i] /= (aec->xPow[i] + 1e-10f);
+    absEf = sqrtf(ef[0][i] * ef[0][i] + ef[1][i] * ef[1][i]);
+
+    if (absEf > aec->errThresh) {
+      absEf = aec->errThresh / (absEf + 1e-10f);
+      ef[0][i] *= absEf;
+      ef[1][i] *= absEf;
+    }
+
+    // Stepsize factor
+    ef[0][i] *= aec->mu;
+    ef[1][i] *= aec->mu;
+  }
+}
+
+static void FilterAdaptationSSE2(aec_t *aec, float *fft, float ef[2][PART_LEN1]) {
+  int i, j;
+  for (i = 0; i < NR_PART; i++) {
+    int xPos = (i + aec->xfBufBlockPos)*(PART_LEN1);
+    int pos = i * PART_LEN1;
+    // Check for wrap
+    if (i + aec->xfBufBlockPos >= NR_PART) {
+      xPos -= NR_PART * PART_LEN1;
+    }
+
+#ifdef UNCONSTR
+    for (j = 0; j < PART_LEN1; j++) {
+      aec->wfBuf[pos + j][0] += MulRe(aec->xfBuf[xPos + j][0],
+                                      -aec->xfBuf[xPos + j][1],
+                                      ef[j][0], ef[j][1]);
+      aec->wfBuf[pos + j][1] += MulIm(aec->xfBuf[xPos + j][0],
+                                      -aec->xfBuf[xPos + j][1],
+                                      ef[j][0], ef[j][1]);
+    }
+#else
+    // Process the whole array...
+    for (j = 0; j < PART_LEN; j+= 4) {
+      // Load xfBuf and ef.
+      const __m128 xfBuf_re = _mm_loadu_ps(&aec->xfBuf[0][xPos + j]);
+      const __m128 xfBuf_im = _mm_loadu_ps(&aec->xfBuf[1][xPos + j]);
+      const __m128 ef_re = _mm_loadu_ps(&ef[0][j]);
+      const __m128 ef_im = _mm_loadu_ps(&ef[1][j]);
+      // Calculate the product of conjugate(xfBuf) by ef.
+      //   re(conjugate(a) * b) = aRe * bRe + aIm * bIm
+      //   im(conjugate(a) * b)=  aRe * bIm - aIm * bRe
+      const __m128 a = _mm_mul_ps(xfBuf_re, ef_re);
+      const __m128 b = _mm_mul_ps(xfBuf_im, ef_im);
+      const __m128 c = _mm_mul_ps(xfBuf_re, ef_im);
+      const __m128 d = _mm_mul_ps(xfBuf_im, ef_re);
+      const __m128 e = _mm_add_ps(a, b);
+      const __m128 f = _mm_sub_ps(c, d);
+      // Interleave real and imaginary parts.
+      const __m128 g = _mm_unpacklo_ps(e, f);
+      const __m128 h = _mm_unpackhi_ps(e, f);
+      // Store
+      _mm_storeu_ps(&fft[2*j + 0], g);
+      _mm_storeu_ps(&fft[2*j + 4], h);
+    }
+    // ... and fixup the first imaginary entry.
+    fft[1] = MulRe(aec->xfBuf[0][xPos + PART_LEN],
+                   -aec->xfBuf[1][xPos + PART_LEN],
+                   ef[0][PART_LEN], ef[1][PART_LEN]);
+
+    aec_rdft_inverse_128(fft);
+    memset(fft + PART_LEN, 0, sizeof(float)*PART_LEN);
+
+    // fft scaling
+    {
+      float scale = 2.0f / PART_LEN2;
+      const __m128 scale_ps = _mm_load_ps1(&scale);
+      for (j = 0; j < PART_LEN; j+=4) {
+        const __m128 fft_ps = _mm_loadu_ps(&fft[j]);
+        const __m128 fft_scale = _mm_mul_ps(fft_ps, scale_ps);
+        _mm_storeu_ps(&fft[j], fft_scale);
+      }
+    }
+    aec_rdft_forward_128(fft);
+
+    {
+      float wt1 = aec->wfBuf[1][pos];
+      aec->wfBuf[0][pos + PART_LEN] += fft[1];
+      for (j = 0; j < PART_LEN; j+= 4) {
+        __m128 wtBuf_re = _mm_loadu_ps(&aec->wfBuf[0][pos + j]);
+        __m128 wtBuf_im = _mm_loadu_ps(&aec->wfBuf[1][pos + j]);
+        const __m128 fft0 = _mm_loadu_ps(&fft[2 * j + 0]);
+        const __m128 fft4 = _mm_loadu_ps(&fft[2 * j + 4]);
+        const __m128 fft_re = _mm_shuffle_ps(fft0, fft4, _MM_SHUFFLE(2, 0, 2 ,0));
+        const __m128 fft_im = _mm_shuffle_ps(fft0, fft4, _MM_SHUFFLE(3, 1, 3 ,1));
+        wtBuf_re = _mm_add_ps(wtBuf_re, fft_re);
+        wtBuf_im = _mm_add_ps(wtBuf_im, fft_im);
+        _mm_storeu_ps(&aec->wfBuf[0][pos + j], wtBuf_re);
+        _mm_storeu_ps(&aec->wfBuf[1][pos + j], wtBuf_im);
+      }
+      aec->wfBuf[1][pos] = wt1;
+    }
+#endif // UNCONSTR
+  }
+}
+
+static __m128 mm_pow_ps(__m128 a, __m128 b)
+{
+  // a^b = exp2(b * log2(a))
+  //   exp2(x) and log2(x) are calculated using polynomial approximations.
+  __m128 log2_a, b_log2_a, a_exp_b;
+
+  // Calculate log2(x), x = a.
+  {
+    // To calculate log2(x), we decompose x like this:
+    //   x = y * 2^n
+    //     n is an integer
+    //     y is in the [1.0, 2.0) range
+    //
+    //   log2(x) = log2(y) + n
+    //     n       can be evaluated by playing with float representation.
+    //     log2(y) in a small range can be approximated, this code uses an order
+    //             five polynomial approximation. The coefficients have been
+    //             estimated with the Remez algorithm and the resulting
+    //             polynomial has a maximum relative error of 0.00086%.
+
+    // Compute n.
+    //    This is done by masking the exponent, shifting it into the top bit of
+    //    the mantissa, putting eight into the biased exponent (to shift/
+    //    compensate the fact that the exponent has been shifted in the top/
+    //    fractional part and finally getting rid of the implicit leading one
+    //    from the mantissa by substracting it out.
+    static const ALIGN16_BEG int float_exponent_mask[4] ALIGN16_END =
+        {0x7F800000, 0x7F800000, 0x7F800000, 0x7F800000};
+    static const ALIGN16_BEG int eight_biased_exponent[4] ALIGN16_END =
+        {0x43800000, 0x43800000, 0x43800000, 0x43800000};
+    static const ALIGN16_BEG int implicit_leading_one[4] ALIGN16_END =
+        {0x43BF8000, 0x43BF8000, 0x43BF8000, 0x43BF8000};
+    static const int shift_exponent_into_top_mantissa = 8;
+    const __m128 two_n = _mm_and_ps(a, *((__m128 *)float_exponent_mask));
+    const __m128 n_1 = (__m128)_mm_srli_epi32((__m128i)two_n,
+        shift_exponent_into_top_mantissa);
+    const __m128 n_0 = _mm_or_ps(
+        (__m128)n_1, *((__m128 *)eight_biased_exponent));
+    const __m128 n   = _mm_sub_ps(n_0,  *((__m128 *)implicit_leading_one));
+
+    // Compute y.
+    static const ALIGN16_BEG int mantissa_mask[4] ALIGN16_END =
+        {0x007FFFFF, 0x007FFFFF, 0x007FFFFF, 0x007FFFFF};
+    static const ALIGN16_BEG int zero_biased_exponent_is_one[4] ALIGN16_END =
+        {0x3F800000, 0x3F800000, 0x3F800000, 0x3F800000};
+    const __m128 mantissa = _mm_and_ps(a, *((__m128 *)mantissa_mask));
+    const __m128 y        = _mm_or_ps(
+        mantissa,  *((__m128 *)zero_biased_exponent_is_one));
+
+    // Approximate log2(y) ~= (y - 1) * pol5(y).
+    //    pol5(y) = C5 * y^5 + C4 * y^4 + C3 * y^3 + C2 * y^2 + C1 * y + C0
+    static const ALIGN16_BEG float ALIGN16_END C5[4] =
+        {-3.4436006e-2f, -3.4436006e-2f, -3.4436006e-2f, -3.4436006e-2f};
+    static const ALIGN16_BEG float ALIGN16_END C4[4] =
+        {3.1821337e-1f, 3.1821337e-1f, 3.1821337e-1f, 3.1821337e-1f};
+    static const ALIGN16_BEG float ALIGN16_END C3[4] =
+        {-1.2315303f, -1.2315303f, -1.2315303f, -1.2315303f};
+    static const ALIGN16_BEG float ALIGN16_END C2[4] =
+        {2.5988452f, 2.5988452f, 2.5988452f, 2.5988452f};
+    static const ALIGN16_BEG float ALIGN16_END C1[4] =
+        {-3.3241990f, -3.3241990f, -3.3241990f, -3.3241990f};
+    static const ALIGN16_BEG float ALIGN16_END C0[4] =
+        {3.1157899f, 3.1157899f, 3.1157899f, 3.1157899f};
+    const __m128 pol5_y_0 = _mm_mul_ps(y,        *((__m128 *)C5));
+    const __m128 pol5_y_1 = _mm_add_ps(pol5_y_0, *((__m128 *)C4));
+    const __m128 pol5_y_2 = _mm_mul_ps(pol5_y_1, y);
+    const __m128 pol5_y_3 = _mm_add_ps(pol5_y_2, *((__m128 *)C3));
+    const __m128 pol5_y_4 = _mm_mul_ps(pol5_y_3, y);
+    const __m128 pol5_y_5 = _mm_add_ps(pol5_y_4, *((__m128 *)C2));
+    const __m128 pol5_y_6 = _mm_mul_ps(pol5_y_5, y);
+    const __m128 pol5_y_7 = _mm_add_ps(pol5_y_6, *((__m128 *)C1));
+    const __m128 pol5_y_8 = _mm_mul_ps(pol5_y_7, y);
+    const __m128 pol5_y   = _mm_add_ps(pol5_y_8, *((__m128 *)C0));
+    const __m128 y_minus_one = _mm_sub_ps(
+        y, *((__m128 *)zero_biased_exponent_is_one));
+    const __m128 log2_y = _mm_mul_ps(y_minus_one ,  pol5_y);
+
+    // Combine parts.
+    log2_a = _mm_add_ps(n, log2_y);
+  }
+
+  // b * log2(a)
+  b_log2_a = _mm_mul_ps(b, log2_a);
+
+  // Calculate exp2(x), x = b * log2(a).
+  {
+    // To calculate 2^x, we decompose x like this:
+    //   x = n + y
+    //     n is an integer, the value of x - 0.5 rounded down, therefore
+    //     y is in the [0.5, 1.5) range
+    //
+    //   2^x = 2^n * 2^y
+    //     2^n can be evaluated by playing with float representation.
+    //     2^y in a small range can be approximated, this code uses an order two
+    //         polynomial approximation. The coefficients have been estimated
+    //         with the Remez algorithm and the resulting polynomial has a
+    //         maximum relative error of 0.17%.
+
+    // To avoid over/underflow, we reduce the range of input to ]-127, 129].
+    static const ALIGN16_BEG float max_input[4] ALIGN16_END =
+        {129.f, 129.f, 129.f, 129.f};
+    static const ALIGN16_BEG float min_input[4] ALIGN16_END =
+        {-126.99999f, -126.99999f, -126.99999f, -126.99999f};
+    const __m128 x_min = _mm_min_ps(b_log2_a, *((__m128 *)max_input));
+    const __m128 x_max = _mm_max_ps(x_min,    *((__m128 *)min_input));
+    // Compute n.
+    static const ALIGN16_BEG float half[4] ALIGN16_END =
+        {0.5f, 0.5f, 0.5f, 0.5f};
+    const __m128  x_minus_half = _mm_sub_ps(x_max, *((__m128 *)half));
+    const __m128i x_minus_half_floor = _mm_cvtps_epi32(x_minus_half);
+    // Compute 2^n.
+    static const ALIGN16_BEG int float_exponent_bias[4] ALIGN16_END =
+        {127, 127, 127, 127};
+    static const int float_exponent_shift = 23;
+    const __m128i two_n_exponent = _mm_add_epi32(
+        x_minus_half_floor, *((__m128i *)float_exponent_bias));
+    const __m128  two_n = (__m128)_mm_slli_epi32(
+        two_n_exponent, float_exponent_shift);
+    // Compute y.
+    const __m128 y = _mm_sub_ps(x_max, _mm_cvtepi32_ps(x_minus_half_floor));
+    // Approximate 2^y ~= C2 * y^2 + C1 * y + C0.
+    static const ALIGN16_BEG float C2[4] ALIGN16_END =
+        {3.3718944e-1f, 3.3718944e-1f, 3.3718944e-1f, 3.3718944e-1f};
+    static const ALIGN16_BEG float C1[4] ALIGN16_END =
+        {6.5763628e-1f, 6.5763628e-1f, 6.5763628e-1f, 6.5763628e-1f};
+    static const ALIGN16_BEG float C0[4] ALIGN16_END =
+        {1.0017247f, 1.0017247f, 1.0017247f, 1.0017247f};
+    const __m128 exp2_y_0 = _mm_mul_ps(y,        *((__m128 *)C2));
+    const __m128 exp2_y_1 = _mm_add_ps(exp2_y_0, *((__m128 *)C1));
+    const __m128 exp2_y_2 = _mm_mul_ps(exp2_y_1, y);
+    const __m128 exp2_y   = _mm_add_ps(exp2_y_2, *((__m128 *)C0));
+
+    // Combine parts.
+    a_exp_b = _mm_mul_ps(exp2_y, two_n);
+  }
+  return a_exp_b;
+}
+
+extern const float WebRtcAec_weightCurve[65];
+extern const float WebRtcAec_overDriveCurve[65];
+
+static void OverdriveAndSuppressSSE2(aec_t *aec, float hNl[PART_LEN1],
+                                     const float hNlFb,
+                                     float efw[2][PART_LEN1]) {
+  int i;
+  const __m128 vec_hNlFb = _mm_set1_ps(hNlFb);
+  const __m128 vec_one = _mm_set1_ps(1.0f);
+  const __m128 vec_minus_one = _mm_set1_ps(-1.0f);
+  const __m128 vec_overDriveSm = _mm_set1_ps(aec->overDriveSm);
+  // vectorized code (four at once)
+  for (i = 0; i + 3 < PART_LEN1; i+=4) {
+    // Weight subbands
+    __m128 vec_hNl = _mm_loadu_ps(&hNl[i]);
+    const __m128 vec_weightCurve = _mm_loadu_ps(&WebRtcAec_weightCurve[i]);
+    const __m128 bigger = _mm_cmpgt_ps(vec_hNl, vec_hNlFb);
+    const __m128 vec_weightCurve_hNlFb = _mm_mul_ps(
+        vec_weightCurve, vec_hNlFb);
+    const __m128 vec_one_weightCurve = _mm_sub_ps(vec_one, vec_weightCurve);
+    const __m128 vec_one_weightCurve_hNl = _mm_mul_ps(
+        vec_one_weightCurve, vec_hNl);
+    const __m128 vec_if0 = _mm_andnot_ps(bigger, vec_hNl);
+    const __m128 vec_if1 = _mm_and_ps(
+        bigger, _mm_add_ps(vec_weightCurve_hNlFb, vec_one_weightCurve_hNl));
+    vec_hNl = _mm_or_ps(vec_if0, vec_if1);
+
+    {
+      const __m128 vec_overDriveCurve = _mm_loadu_ps(
+          &WebRtcAec_overDriveCurve[i]);
+      const __m128 vec_overDriveSm_overDriveCurve = _mm_mul_ps(
+          vec_overDriveSm, vec_overDriveCurve);
+      vec_hNl = mm_pow_ps(vec_hNl, vec_overDriveSm_overDriveCurve);
+      _mm_storeu_ps(&hNl[i], vec_hNl);
+    }
+
+    // Suppress error signal
+    {
+      __m128 vec_efw_re = _mm_loadu_ps(&efw[0][i]);
+      __m128 vec_efw_im = _mm_loadu_ps(&efw[1][i]);
+      vec_efw_re = _mm_mul_ps(vec_efw_re, vec_hNl);
+      vec_efw_im = _mm_mul_ps(vec_efw_im, vec_hNl);
+
+      // Ooura fft returns incorrect sign on imaginary component. It matters
+      // here because we are making an additive change with comfort noise.
+      vec_efw_im = _mm_mul_ps(vec_efw_im, vec_minus_one);
+      _mm_storeu_ps(&efw[0][i], vec_efw_re);
+      _mm_storeu_ps(&efw[1][i], vec_efw_im);
+    }
+  }
+  // scalar code for the remaining items.
+  for (; i < PART_LEN1; i++) {
+    // Weight subbands
+    if (hNl[i] > hNlFb) {
+      hNl[i] = WebRtcAec_weightCurve[i] * hNlFb +
+          (1 - WebRtcAec_weightCurve[i]) * hNl[i];
+    }
+    hNl[i] = powf(hNl[i], aec->overDriveSm * WebRtcAec_overDriveCurve[i]);
+
+    // Suppress error signal
+    efw[0][i] *= hNl[i];
+    efw[1][i] *= hNl[i];
+
+    // Ooura fft returns incorrect sign on imaginary component. It matters
+    // here because we are making an additive change with comfort noise.
+    efw[1][i] *= -1;
+  }
+}
+
+void WebRtcAec_InitAec_SSE2(void) {
+  WebRtcAec_FilterFar = FilterFarSSE2;
+  WebRtcAec_ScaleErrorSignal = ScaleErrorSignalSSE2;
+  WebRtcAec_FilterAdaptation = FilterAdaptationSSE2;
+  WebRtcAec_OverdriveAndSuppress = OverdriveAndSuppressSSE2;
+}
+
+#endif   // WEBRTC_USE_SSE2

src/modules/audio_processing/aec/main/source/aec_rdft.c

@@ -0,0 +1,587 @@
+/*
+ * http://www.kurims.kyoto-u.ac.jp/~ooura/fft.html
+ * Copyright Takuya OOURA, 1996-2001
+ *
+ * You may use, copy, modify and distribute this code for any purpose (include
+ * commercial use) and without fee. Please refer to this package when you modify
+ * this code.
+ *
+ * Changes by the WebRTC authors:
+ *    - Trivial type modifications.
+ *    - Minimal code subset to do rdft of length 128.
+ *    - Optimizations because of known length.
+ *
+ *  All changes are covered by the WebRTC license and IP grant:
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "aec_rdft.h"
+
+#include <math.h>
+
+#include "system_wrappers/interface/cpu_features_wrapper.h"
+#include "typedefs.h"
+
+// constants shared by all paths (C, SSE2).
+float rdft_w[64];
+// constants used by the C path.
+float rdft_wk3ri_first[32];
+float rdft_wk3ri_second[32];
+// constants used by SSE2 but initialized in C path.
+ALIGN16_BEG float ALIGN16_END rdft_wk1r[32];
+ALIGN16_BEG float ALIGN16_END rdft_wk2r[32];
+ALIGN16_BEG float ALIGN16_END rdft_wk3r[32];
+ALIGN16_BEG float ALIGN16_END rdft_wk1i[32];
+ALIGN16_BEG float ALIGN16_END rdft_wk2i[32];
+ALIGN16_BEG float ALIGN16_END rdft_wk3i[32];
+ALIGN16_BEG float ALIGN16_END cftmdl_wk1r[4];
+
+static int ip[16];
+
+static void bitrv2_32or128(int n, int *ip, float *a) {
+  // n is 32 or 128
+  int j, j1, k, k1, m, m2;
+  float xr, xi, yr, yi;
+
+  ip[0] = 0;
+  {
+    int l = n;
+    m = 1;
+    while ((m << 3) < l) {
+      l >>= 1;
+      for (j = 0; j < m; j++) {
+        ip[m + j] = ip[j] + l;
+      }
+      m <<= 1;
+    }
+  }
+  m2 = 2 * m;
+  for (k = 0; k < m; k++) {
+    for (j = 0; j < k; j++) {
+      j1 = 2 * j + ip[k];
+      k1 = 2 * k + ip[j];
+      xr = a[j1];
+      xi = a[j1 + 1];
+      yr = a[k1];
+      yi = a[k1 + 1];
+      a[j1] = yr;
+      a[j1 + 1] = yi;
+      a[k1] = xr;
+      a[k1 + 1] = xi;
+      j1 += m2;
+      k1 += 2 * m2;
+      xr = a[j1];
+      xi = a[j1 + 1];
+      yr = a[k1];
+      yi = a[k1 + 1];
+      a[j1] = yr;
+      a[j1 + 1] = yi;
+      a[k1] = xr;
+      a[k1 + 1] = xi;
+      j1 += m2;
+      k1 -= m2;
+      xr = a[j1];
+      xi = a[j1 + 1];
+      yr = a[k1];
+      yi = a[k1 + 1];
+      a[j1] = yr;
+      a[j1 + 1] = yi;
+      a[k1] = xr;
+      a[k1 + 1] = xi;
+      j1 += m2;
+      k1 += 2 * m2;
+      xr = a[j1];
+      xi = a[j1 + 1];
+      yr = a[k1];
+      yi = a[k1 + 1];
+      a[j1] = yr;
+      a[j1 + 1] = yi;
+      a[k1] = xr;
+      a[k1 + 1] = xi;
+    }
+    j1 = 2 * k + m2 + ip[k];
+    k1 = j1 + m2;
+    xr = a[j1];
+    xi = a[j1 + 1];
+    yr = a[k1];
+    yi = a[k1 + 1];
+    a[j1] = yr;
+    a[j1 + 1] = yi;
+    a[k1] = xr;
+    a[k1 + 1] = xi;
+  }
+}
+
+static void makewt_32(void) {
+  const int nw = 32;
+  int j, nwh;
+  float delta, x, y;
+
+  ip[0] = nw;
+  ip[1] = 1;
+  nwh = nw >> 1;
+  delta = atanf(1.0f) / nwh;
+  rdft_w[0] = 1;
+  rdft_w[1] = 0;
+  rdft_w[nwh] = cosf(delta * nwh);
+  rdft_w[nwh + 1] = rdft_w[nwh];
+  for (j = 2; j < nwh; j += 2) {
+    x = cosf(delta * j);
+    y = sinf(delta * j);
+    rdft_w[j] = x;
+    rdft_w[j + 1] = y;
+    rdft_w[nw - j] = y;
+    rdft_w[nw - j + 1] = x;
+  }
+  bitrv2_32or128(nw, ip + 2, rdft_w);
+
+  // pre-calculate constants used by cft1st_128 and cftmdl_128...
+  cftmdl_wk1r[0] = rdft_w[2];
+  cftmdl_wk1r[1] = rdft_w[2];
+  cftmdl_wk1r[2] = rdft_w[2];
+  cftmdl_wk1r[3] = -rdft_w[2];
+  {
+    int k1;
+
+    for (k1 = 0, j = 0; j < 128; j += 16, k1 += 2) {
+      const int k2 = 2 * k1;
+      const float wk2r = rdft_w[k1 + 0];
+      const float wk2i = rdft_w[k1 + 1];
+      float wk1r, wk1i;
+      // ... scalar version.
+      wk1r = rdft_w[k2 + 0];
+      wk1i = rdft_w[k2 + 1];
+      rdft_wk3ri_first[k1 + 0] = wk1r - 2 * wk2i * wk1i;
+      rdft_wk3ri_first[k1 + 1] = 2 * wk2i * wk1r - wk1i;
+      wk1r = rdft_w[k2 + 2];
+      wk1i = rdft_w[k2 + 3];
+      rdft_wk3ri_second[k1 + 0] = wk1r - 2 * wk2r * wk1i;
+      rdft_wk3ri_second[k1 + 1] = 2 * wk2r * wk1r - wk1i;
+      // ... vector version.
+      rdft_wk1r[k2 + 0] = rdft_w[k2 + 0];
+      rdft_wk1r[k2 + 1] = rdft_w[k2 + 0];
+      rdft_wk1r[k2 + 2] = rdft_w[k2 + 2];
+      rdft_wk1r[k2 + 3] = rdft_w[k2 + 2];
+      rdft_wk2r[k2 + 0] = rdft_w[k1 + 0];
+      rdft_wk2r[k2 + 1] = rdft_w[k1 + 0];
+      rdft_wk2r[k2 + 2] = -rdft_w[k1 + 1];
+      rdft_wk2r[k2 + 3] = -rdft_w[k1 + 1];
+      rdft_wk3r[k2 + 0] = rdft_wk3ri_first[k1 + 0];
+      rdft_wk3r[k2 + 1] = rdft_wk3ri_first[k1 + 0];
+      rdft_wk3r[k2 + 2] = rdft_wk3ri_second[k1 + 0];
+      rdft_wk3r[k2 + 3] = rdft_wk3ri_second[k1 + 0];
+      rdft_wk1i[k2 + 0] = -rdft_w[k2 + 1];
+      rdft_wk1i[k2 + 1] = rdft_w[k2 + 1];
+      rdft_wk1i[k2 + 2] = -rdft_w[k2 + 3];
+      rdft_wk1i[k2 + 3] = rdft_w[k2 + 3];
+      rdft_wk2i[k2 + 0] = -rdft_w[k1 + 1];
+      rdft_wk2i[k2 + 1] = rdft_w[k1 + 1];
+      rdft_wk2i[k2 + 2] = -rdft_w[k1 + 0];
+      rdft_wk2i[k2 + 3] = rdft_w[k1 + 0];
+      rdft_wk3i[k2 + 0] = -rdft_wk3ri_first[k1 + 1];
+      rdft_wk3i[k2 + 1] = rdft_wk3ri_first[k1 + 1];
+      rdft_wk3i[k2 + 2] = -rdft_wk3ri_second[k1 + 1];
+      rdft_wk3i[k2 + 3] = rdft_wk3ri_second[k1 + 1];
+    }
+  }
+}
+
+static void makect_32(void) {
+  float *c = rdft_w + 32;
+  const int nc = 32;
+  int j, nch;
+  float delta;
+
+  ip[1] = nc;
+  nch = nc >> 1;
+  delta = atanf(1.0f) / nch;
+  c[0] = cosf(delta * nch);
+  c[nch] = 0.5f * c[0];
+  for (j = 1; j < nch; j++) {
+    c[j] = 0.5f * cosf(delta * j);
+    c[nc - j] = 0.5f * sinf(delta * j);
+  }
+}
+
+static void cft1st_128_C(float *a) {
+  const int n = 128;
+  int j, k1, k2;
+  float wk1r, wk1i, wk2r, wk2i, wk3r, wk3i;
+  float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;
+
+  x0r = a[0] + a[2];
+  x0i = a[1] + a[3];
+  x1r = a[0] - a[2];
+  x1i = a[1] - a[3];
+  x2r = a[4] + a[6];
+  x2i = a[5] + a[7];
+  x3r = a[4] - a[6];
+  x3i = a[5] - a[7];
+  a[0] = x0r + x2r;
+  a[1] = x0i + x2i;
+  a[4] = x0r - x2r;
+  a[5] = x0i - x2i;
+  a[2] = x1r - x3i;
+  a[3] = x1i + x3r;
+  a[6] = x1r + x3i;
+  a[7] = x1i - x3r;
+  wk1r = rdft_w[2];
+  x0r = a[8] + a[10];
+  x0i = a[9] + a[11];
+  x1r = a[8] - a[10];
+  x1i = a[9] - a[11];
+  x2r = a[12] + a[14];
+  x2i = a[13] + a[15];
+  x3r = a[12] - a[14];
+  x3i = a[13] - a[15];
+  a[8] = x0r + x2r;
+  a[9] = x0i + x2i;
+  a[12] = x2i - x0i;
+  a[13] = x0r - x2r;
+  x0r = x1r - x3i;
+  x0i = x1i + x3r;
+  a[10] = wk1r * (x0r - x0i);
+  a[11] = wk1r * (x0r + x0i);
+  x0r = x3i + x1r;
+  x0i = x3r - x1i;
+  a[14] = wk1r * (x0i - x0r);
+  a[15] = wk1r * (x0i + x0r);
+  k1 = 0;
+  for (j = 16; j < n; j += 16) {
+    k1 += 2;
+    k2 = 2 * k1;
+    wk2r = rdft_w[k1 + 0];
+    wk2i = rdft_w[k1 + 1];
+    wk1r = rdft_w[k2 + 0];
+    wk1i = rdft_w[k2 + 1];
+    wk3r = rdft_wk3ri_first[k1 + 0];
+    wk3i = rdft_wk3ri_first[k1 + 1];
+    x0r = a[j + 0] + a[j + 2];
+    x0i = a[j + 1] + a[j + 3];
+    x1r = a[j + 0] - a[j + 2];
+    x1i = a[j + 1] - a[j + 3];
+    x2r = a[j + 4] + a[j + 6];
+    x2i = a[j + 5] + a[j + 7];
+    x3r = a[j + 4] - a[j + 6];
+    x3i = a[j + 5] - a[j + 7];
+    a[j + 0] = x0r + x2r;
+    a[j + 1] = x0i + x2i;
+    x0r -= x2r;
+    x0i -= x2i;
+    a[j + 4] = wk2r * x0r - wk2i * x0i;
+    a[j + 5] = wk2r * x0i + wk2i * x0r;
+    x0r = x1r - x3i;
+    x0i = x1i + x3r;
+    a[j + 2] = wk1r * x0r - wk1i * x0i;
+    a[j + 3] = wk1r * x0i + wk1i * x0r;
+    x0r = x1r + x3i;
+    x0i = x1i - x3r;
+    a[j + 6] = wk3r * x0r - wk3i * x0i;
+    a[j + 7] = wk3r * x0i + wk3i * x0r;
+    wk1r = rdft_w[k2 + 2];
+    wk1i = rdft_w[k2 + 3];
+    wk3r = rdft_wk3ri_second[k1 + 0];
+    wk3i = rdft_wk3ri_second[k1 + 1];
+    x0r = a[j + 8] + a[j + 10];
+    x0i = a[j + 9] + a[j + 11];
+    x1r = a[j + 8] - a[j + 10];
+    x1i = a[j + 9] - a[j + 11];
+    x2r = a[j + 12] + a[j + 14];
+    x2i = a[j + 13] + a[j + 15];
+    x3r = a[j + 12] - a[j + 14];
+    x3i = a[j + 13] - a[j + 15];
+    a[j + 8] = x0r + x2r;
+    a[j + 9] = x0i + x2i;
+    x0r -= x2r;
+    x0i -= x2i;
+    a[j + 12] = -wk2i * x0r - wk2r * x0i;
+    a[j + 13] = -wk2i * x0i + wk2r * x0r;
+    x0r = x1r - x3i;
+    x0i = x1i + x3r;
+    a[j + 10] = wk1r * x0r - wk1i * x0i;
+    a[j + 11] = wk1r * x0i + wk1i * x0r;
+    x0r = x1r + x3i;
+    x0i = x1i - x3r;
+    a[j + 14] = wk3r * x0r - wk3i * x0i;
+    a[j + 15] = wk3r * x0i + wk3i * x0r;
+  }
+}
+
+static void cftmdl_128_C(float *a) {
+  const int l = 8;
+  const int n = 128;
+  const int m = 32;
+  int j0, j1, j2, j3, k, k1, k2, m2;
+  float wk1r, wk1i, wk2r, wk2i, wk3r, wk3i;
+  float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;
+
+  for (j0 = 0; j0 < l; j0 += 2) {
+    j1 = j0 +  8;
+    j2 = j0 + 16;
+    j3 = j0 + 24;
+    x0r = a[j0 + 0] + a[j1 + 0];
+    x0i = a[j0 + 1] + a[j1 + 1];
+    x1r = a[j0 + 0] - a[j1 + 0];
+    x1i = a[j0 + 1] - a[j1 + 1];
+    x2r = a[j2 + 0] + a[j3 + 0];
+    x2i = a[j2 + 1] + a[j3 + 1];
+    x3r = a[j2 + 0] - a[j3 + 0];
+    x3i = a[j2 + 1] - a[j3 + 1];
+    a[j0 + 0] = x0r + x2r;
+    a[j0 + 1] = x0i + x2i;
+    a[j2 + 0] = x0r - x2r;
+    a[j2 + 1] = x0i - x2i;
+    a[j1 + 0] = x1r - x3i;
+    a[j1 + 1] = x1i + x3r;
+    a[j3 + 0] = x1r + x3i;
+    a[j3 + 1] = x1i - x3r;
+  }
+  wk1r = rdft_w[2];
+  for (j0 = m; j0 < l + m; j0 += 2) {
+    j1 = j0 +  8;
+    j2 = j0 + 16;
+    j3 = j0 + 24;
+    x0r = a[j0 + 0] + a[j1 + 0];
+    x0i = a[j0 + 1] + a[j1 + 1];
+    x1r = a[j0 + 0] - a[j1 + 0];
+    x1i = a[j0 + 1] - a[j1 + 1];
+    x2r = a[j2 + 0] + a[j3 + 0];
+    x2i = a[j2 + 1] + a[j3 + 1];
+    x3r = a[j2 + 0] - a[j3 + 0];
+    x3i = a[j2 + 1] - a[j3 + 1];
+    a[j0 + 0] = x0r + x2r;
+    a[j0 + 1] = x0i + x2i;
+    a[j2 + 0] = x2i - x0i;
+    a[j2 + 1] = x0r - x2r;
+    x0r = x1r - x3i;
+    x0i = x1i + x3r;
+    a[j1 + 0] = wk1r * (x0r - x0i);
+    a[j1 + 1] = wk1r * (x0r + x0i);
+    x0r = x3i + x1r;
+    x0i = x3r - x1i;
+    a[j3 + 0] = wk1r * (x0i - x0r);
+    a[j3 + 1] = wk1r * (x0i + x0r);
+  }
+  k1 = 0;
+  m2 = 2 * m;
+  for (k = m2; k < n; k += m2) {
+    k1 += 2;
+    k2 = 2 * k1;
+    wk2r = rdft_w[k1 + 0];
+    wk2i = rdft_w[k1 + 1];
+    wk1r = rdft_w[k2 + 0];
+    wk1i = rdft_w[k2 + 1];
+    wk3r = rdft_wk3ri_first[k1 + 0];
+    wk3i = rdft_wk3ri_first[k1 + 1];
+    for (j0 = k; j0 < l + k; j0 += 2) {
+      j1 = j0 +  8;
+      j2 = j0 + 16;
+      j3 = j0 + 24;
+      x0r = a[j0 + 0] + a[j1 + 0];
+      x0i = a[j0 + 1] + a[j1 + 1];
+      x1r = a[j0 + 0] - a[j1 + 0];
+      x1i = a[j0 + 1] - a[j1 + 1];
+      x2r = a[j2 + 0] + a[j3 + 0];
+      x2i = a[j2 + 1] + a[j3 + 1];
+      x3r = a[j2 + 0] - a[j3 + 0];
+      x3i = a[j2 + 1] - a[j3 + 1];
+      a[j0 + 0] = x0r + x2r;
+      a[j0 + 1] = x0i + x2i;
+      x0r -= x2r;
+      x0i -= x2i;
+      a[j2 + 0] = wk2r * x0r - wk2i * x0i;
+      a[j2 + 1] = wk2r * x0i + wk2i * x0r;
+      x0r = x1r - x3i;
+      x0i = x1i + x3r;
+      a[j1 + 0] = wk1r * x0r - wk1i * x0i;
+      a[j1 + 1] = wk1r * x0i + wk1i * x0r;
+      x0r = x1r + x3i;
+      x0i = x1i - x3r;
+      a[j3 + 0] = wk3r * x0r - wk3i * x0i;
+      a[j3 + 1] = wk3r * x0i + wk3i * x0r;
+    }
+    wk1r = rdft_w[k2 + 2];
+    wk1i = rdft_w[k2 + 3];
+    wk3r = rdft_wk3ri_second[k1 + 0];
+    wk3i = rdft_wk3ri_second[k1 + 1];
+    for (j0 = k + m; j0 < l + (k + m); j0 += 2) {
+      j1 = j0 +  8;
+      j2 = j0 + 16;
+      j3 = j0 + 24;
+      x0r = a[j0 + 0] + a[j1 + 0];
+      x0i = a[j0 + 1] + a[j1 + 1];
+      x1r = a[j0 + 0] - a[j1 + 0];
+      x1i = a[j0 + 1] - a[j1 + 1];
+      x2r = a[j2 + 0] + a[j3 + 0];
+      x2i = a[j2 + 1] + a[j3 + 1];
+      x3r = a[j2 + 0] - a[j3 + 0];
+      x3i = a[j2 + 1] - a[j3 + 1];
+      a[j0 + 0] = x0r + x2r;
+      a[j0 + 1] = x0i + x2i;
+      x0r -= x2r;
+      x0i -= x2i;
+      a[j2 + 0] = -wk2i * x0r - wk2r * x0i;
+      a[j2 + 1] = -wk2i * x0i + wk2r * x0r;
+      x0r = x1r - x3i;
+      x0i = x1i + x3r;
+      a[j1 + 0] = wk1r * x0r - wk1i * x0i;
+      a[j1 + 1] = wk1r * x0i + wk1i * x0r;
+      x0r = x1r + x3i;
+      x0i = x1i - x3r;
+      a[j3 + 0] = wk3r * x0r - wk3i * x0i;
+      a[j3 + 1] = wk3r * x0i + wk3i * x0r;
+    }
+  }
+}
+
+static void cftfsub_128(float *a) {
+  int j, j1, j2, j3, l;
+  float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;
+
+  cft1st_128(a);
+  cftmdl_128(a);
+  l = 32;
+  for (j = 0; j < l; j += 2) {
+    j1 = j + l;
+    j2 = j1 + l;
+    j3 = j2 + l;
+    x0r = a[j] + a[j1];
+    x0i = a[j + 1] + a[j1 + 1];
+    x1r = a[j] - a[j1];
+    x1i = a[j + 1] - a[j1 + 1];
+    x2r = a[j2] + a[j3];
+    x2i = a[j2 + 1] + a[j3 + 1];
+    x3r = a[j2] - a[j3];
+    x3i = a[j2 + 1] - a[j3 + 1];
+    a[j] = x0r + x2r;
+    a[j + 1] = x0i + x2i;
+    a[j2] = x0r - x2r;
+    a[j2 + 1] = x0i - x2i;
+    a[j1] = x1r - x3i;
+    a[j1 + 1] = x1i + x3r;
+    a[j3] = x1r + x3i;
+    a[j3 + 1] = x1i - x3r;
+  }
+}
+
+static void cftbsub_128(float *a) {
+  int j, j1, j2, j3, l;
+  float x0r, x0i, x1r, x1i, x2r, x2i, x3r, x3i;
+
+  cft1st_128(a);
+  cftmdl_128(a);
+  l = 32;
+
+  for (j = 0; j < l; j += 2) {
+    j1 = j + l;
+    j2 = j1 + l;
+    j3 = j2 + l;
+    x0r = a[j] + a[j1];
+    x0i = -a[j + 1] - a[j1 + 1];
+    x1r = a[j] - a[j1];
+    x1i = -a[j + 1] + a[j1 + 1];
+    x2r = a[j2] + a[j3];
+    x2i = a[j2 + 1] + a[j3 + 1];
+    x3r = a[j2] - a[j3];
+    x3i = a[j2 + 1] - a[j3 + 1];
+    a[j] = x0r + x2r;
+    a[j + 1] = x0i - x2i;
+    a[j2] = x0r - x2r;
+    a[j2 + 1] = x0i + x2i;
+    a[j1] = x1r - x3i;
+    a[j1 + 1] = x1i - x3r;
+    a[j3] = x1r + x3i;
+    a[j3 + 1] = x1i + x3r;
+  }
+}
+
+static void rftfsub_128_C(float *a) {
+  const float *c = rdft_w + 32;
+  int j1, j2, k1, k2;
+  float wkr, wki, xr, xi, yr, yi;
+
+  for (j1 = 1, j2 = 2; j2 < 64; j1 += 1, j2 += 2) {
+    k2 = 128 - j2;
+    k1 =  32 - j1;
+    wkr = 0.5f - c[k1];
+    wki = c[j1];
+    xr = a[j2 + 0] - a[k2 + 0];
+    xi = a[j2 + 1] + a[k2 + 1];
+    yr = wkr * xr - wki * xi;
+    yi = wkr * xi + wki * xr;
+    a[j2 + 0] -= yr;
+    a[j2 + 1] -= yi;
+    a[k2 + 0] += yr;
+    a[k2 + 1] -= yi;
+  }
+}
+
+static void rftbsub_128_C(float *a) {
+  const float *c = rdft_w + 32;
+  int j1, j2, k1, k2;
+  float wkr, wki, xr, xi, yr, yi;
+
+  a[1] = -a[1];
+  for (j1 = 1, j2 = 2; j2 < 64; j1 += 1, j2 += 2) {
+    k2 = 128 - j2;
+    k1 =  32 - j1;
+    wkr = 0.5f - c[k1];
+    wki = c[j1];
+    xr = a[j2 + 0] - a[k2 + 0];
+    xi = a[j2 + 1] + a[k2 + 1];
+    yr = wkr * xr + wki * xi;
+    yi = wkr * xi - wki * xr;
+    a[j2 + 0] = a[j2 + 0] - yr;
+    a[j2 + 1] = yi - a[j2 + 1];
+    a[k2 + 0] = yr + a[k2 + 0];
+    a[k2 + 1] = yi - a[k2 + 1];
+  }
+  a[65] = -a[65];
+}
+
+void aec_rdft_forward_128(float *a) {
+  const int n = 128;
+  float xi;
+
+  bitrv2_32or128(n, ip + 2, a);
+  cftfsub_128(a);
+  rftfsub_128(a);
+  xi = a[0] - a[1];
+  a[0] += a[1];
+  a[1] = xi;
+}
+
+void aec_rdft_inverse_128(float *a) {
+  const int n = 128;
+
+  a[1] = 0.5f * (a[0] - a[1]);
+  a[0] -= a[1];
+  rftbsub_128(a);
+  bitrv2_32or128(n, ip + 2, a);
+  cftbsub_128(a);
+}
+
+// code path selection
+rft_sub_128_t cft1st_128;
+rft_sub_128_t cftmdl_128;
+rft_sub_128_t rftfsub_128;
+rft_sub_128_t rftbsub_128;
+
+void aec_rdft_init(void) {
+  cft1st_128 = cft1st_128_C;
+  cftmdl_128 = cftmdl_128_C;
+  rftfsub_128 = rftfsub_128_C;
+  rftbsub_128 = rftbsub_128_C;
+  if (WebRtc_GetCPUInfo(kSSE2)) {
+#if defined(WEBRTC_USE_SSE2)
+    aec_rdft_init_sse2();
+#endif
+  }
+  // init library constants.
+  makewt_32();
+  makect_32();
+}

src/modules/audio_processing/aec/main/source/aec_rdft.h

@@ -0,0 +1,49 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_AEC_RDFT_H_
+#define WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_AEC_RDFT_H_
+
+#ifdef _MSC_VER /* visual c++ */
+# define ALIGN16_BEG __declspec(align(16))
+# define ALIGN16_END
+#else /* gcc or icc */
+# define ALIGN16_BEG
+# define ALIGN16_END __attribute__((aligned(16)))
+#endif
+
+// constants shared by all paths (C, SSE2).
+extern float rdft_w[64];
+// constants used by the C path.
+extern float rdft_wk3ri_first[32];
+extern float rdft_wk3ri_second[32];
+// constants used by SSE2 but initialized in C path.
+extern float rdft_wk1r[32];
+extern float rdft_wk2r[32];
+extern float rdft_wk3r[32];
+extern float rdft_wk1i[32];
+extern float rdft_wk2i[32];
+extern float rdft_wk3i[32];
+extern float cftmdl_wk1r[4];
+
+// code path selection function pointers
+typedef void (*rft_sub_128_t)(float *a);
+extern rft_sub_128_t rftfsub_128;
+extern rft_sub_128_t rftbsub_128;
+extern rft_sub_128_t cft1st_128;
+extern rft_sub_128_t cftmdl_128;
+
+// entry points
+void aec_rdft_init(void);
+void aec_rdft_init_sse2(void);
+void aec_rdft_forward_128(float *a);
+void aec_rdft_inverse_128(float *a);
+
+#endif  // WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_AEC_RDFT_H_

src/modules/audio_processing/aec/main/source/aec_rdft_sse2.c

@@ -0,0 +1,397 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "typedefs.h"
+
+#if defined(WEBRTC_USE_SSE2)
+#include <emmintrin.h>
+
+#include "aec_rdft.h"
+
+static const ALIGN16_BEG float ALIGN16_END k_swap_sign[4] =
+  {-1.f, 1.f, -1.f, 1.f};
+
+static void cft1st_128_SSE2(float *a) {
+  const __m128 mm_swap_sign = _mm_load_ps(k_swap_sign);
+  int j, k2;
+
+  for (k2 = 0, j = 0; j < 128; j += 16, k2 += 4) {
+          __m128 a00v   = _mm_loadu_ps(&a[j +  0]);
+          __m128 a04v   = _mm_loadu_ps(&a[j +  4]);
+          __m128 a08v   = _mm_loadu_ps(&a[j +  8]);
+          __m128 a12v   = _mm_loadu_ps(&a[j + 12]);
+          __m128 a01v   = _mm_shuffle_ps(a00v, a08v, _MM_SHUFFLE(1, 0, 1 ,0));
+          __m128 a23v   = _mm_shuffle_ps(a00v, a08v, _MM_SHUFFLE(3, 2, 3 ,2));
+          __m128 a45v   = _mm_shuffle_ps(a04v, a12v, _MM_SHUFFLE(1, 0, 1 ,0));
+          __m128 a67v   = _mm_shuffle_ps(a04v, a12v, _MM_SHUFFLE(3, 2, 3 ,2));
+
+    const __m128 wk1rv  = _mm_load_ps(&rdft_wk1r[k2]);
+    const __m128 wk1iv  = _mm_load_ps(&rdft_wk1i[k2]);
+    const __m128 wk2rv  = _mm_load_ps(&rdft_wk2r[k2]);
+    const __m128 wk2iv  = _mm_load_ps(&rdft_wk2i[k2]);
+    const __m128 wk3rv  = _mm_load_ps(&rdft_wk3r[k2]);
+    const __m128 wk3iv  = _mm_load_ps(&rdft_wk3i[k2]);
+          __m128 x0v    = _mm_add_ps(a01v, a23v);
+    const __m128 x1v    = _mm_sub_ps(a01v, a23v);
+    const __m128 x2v    = _mm_add_ps(a45v, a67v);
+    const __m128 x3v    = _mm_sub_ps(a45v, a67v);
+                 a01v   = _mm_add_ps(x0v, x2v);
+                 x0v    = _mm_sub_ps(x0v, x2v);
+          __m128 x0w    = _mm_shuffle_ps(x0v, x0v, _MM_SHUFFLE(2, 3, 0 ,1));
+
+    const __m128 a45_0v = _mm_mul_ps(wk2rv, x0v);
+    const __m128 a45_1v = _mm_mul_ps(wk2iv, x0w);
+                 a45v   = _mm_add_ps(a45_0v, a45_1v);
+
+    const __m128 x3w    = _mm_shuffle_ps(x3v, x3v, _MM_SHUFFLE(2, 3, 0 ,1));
+    const __m128 x3s    = _mm_mul_ps(mm_swap_sign, x3w);
+                 x0v    = _mm_add_ps(x1v, x3s);
+                 x0w    = _mm_shuffle_ps(x0v, x0v, _MM_SHUFFLE(2, 3, 0 ,1));
+    const __m128 a23_0v = _mm_mul_ps(wk1rv, x0v);
+    const __m128 a23_1v = _mm_mul_ps(wk1iv, x0w);
+                 a23v   = _mm_add_ps(a23_0v, a23_1v);
+
+                 x0v    = _mm_sub_ps(x1v, x3s);
+                 x0w    = _mm_shuffle_ps(x0v, x0v, _MM_SHUFFLE(2, 3, 0 ,1));
+    const __m128 a67_0v = _mm_mul_ps(wk3rv, x0v);
+    const __m128 a67_1v = _mm_mul_ps(wk3iv, x0w);
+                 a67v   = _mm_add_ps(a67_0v, a67_1v);
+
+                 a00v   = _mm_shuffle_ps(a01v, a23v, _MM_SHUFFLE(1, 0, 1 ,0));
+                 a04v   = _mm_shuffle_ps(a45v, a67v, _MM_SHUFFLE(1, 0, 1 ,0));
+                 a08v   = _mm_shuffle_ps(a01v, a23v, _MM_SHUFFLE(3, 2, 3 ,2));
+                 a12v   = _mm_shuffle_ps(a45v, a67v, _MM_SHUFFLE(3, 2, 3 ,2));
+    _mm_storeu_ps(&a[j +  0], a00v);
+    _mm_storeu_ps(&a[j +  4], a04v);
+    _mm_storeu_ps(&a[j +  8], a08v);
+    _mm_storeu_ps(&a[j + 12], a12v);
+  }
+}
+
+static void cftmdl_128_SSE2(float *a) {
+  const int l = 8;
+  const __m128 mm_swap_sign = _mm_load_ps(k_swap_sign);
+  int j0, k, k1, k2;
+
+  __m128 wk1rv = _mm_load_ps(cftmdl_wk1r);
+  for (j0 = 0; j0 < l; j0 += 2) {
+    const __m128i a_00 = _mm_loadl_epi64((__m128i*)&a[j0 +  0]);
+    const __m128i a_08 = _mm_loadl_epi64((__m128i*)&a[j0 +  8]);
+    const __m128i a_32 = _mm_loadl_epi64((__m128i*)&a[j0 + 32]);
+    const __m128i a_40 = _mm_loadl_epi64((__m128i*)&a[j0 + 40]);
+    const __m128  a_00_32 = _mm_shuffle_ps((__m128)a_00, (__m128)a_32,
+                                           _MM_SHUFFLE(1, 0, 1 ,0));
+    const __m128  a_08_40 = _mm_shuffle_ps((__m128)a_08, (__m128)a_40,
+                                           _MM_SHUFFLE(1, 0, 1 ,0));
+          __m128  x0r0_0i0_0r1_x0i1 = _mm_add_ps(a_00_32, a_08_40);
+    const __m128  x1r0_1i0_1r1_x1i1 = _mm_sub_ps(a_00_32, a_08_40);
+
+    const __m128i a_16 = _mm_loadl_epi64((__m128i*)&a[j0 + 16]);
+    const __m128i a_24 = _mm_loadl_epi64((__m128i*)&a[j0 + 24]);
+    const __m128i a_48 = _mm_loadl_epi64((__m128i*)&a[j0 + 48]);
+    const __m128i a_56 = _mm_loadl_epi64((__m128i*)&a[j0 + 56]);
+    const __m128  a_16_48 = _mm_shuffle_ps((__m128)a_16, (__m128)a_48,
+                                           _MM_SHUFFLE(1, 0, 1 ,0));
+    const __m128  a_24_56 = _mm_shuffle_ps((__m128)a_24, (__m128)a_56,
+                                           _MM_SHUFFLE(1, 0, 1 ,0));
+    const __m128  x2r0_2i0_2r1_x2i1 = _mm_add_ps(a_16_48, a_24_56);
+    const __m128  x3r0_3i0_3r1_x3i1 = _mm_sub_ps(a_16_48, a_24_56);
+
+    const __m128  xx0 = _mm_add_ps(x0r0_0i0_0r1_x0i1, x2r0_2i0_2r1_x2i1);
+    _mm_storel_epi64((__m128i*)&a[j0 +  0], (__m128i)xx0);
+    _mm_storel_epi64((__m128i*)&a[j0 + 32],
+                     _mm_shuffle_epi32((__m128i)xx0, _MM_SHUFFLE(3, 2, 3, 2)));
+    const __m128  xx1 = _mm_sub_ps(x0r0_0i0_0r1_x0i1, x2r0_2i0_2r1_x2i1);
+    _mm_storel_epi64((__m128i*)&a[j0 + 16], (__m128i)xx1);
+    _mm_storel_epi64((__m128i*)&a[j0 + 48],
+                     _mm_shuffle_epi32((__m128i)xx1, _MM_SHUFFLE(2, 3, 2, 3)));
+    a[j0 + 48] = -a[j0 + 48];
+
+    const __m128  x3i0_3r0_3i1_x3r1 = (__m128)
+        _mm_shuffle_epi32((__m128i)x3r0_3i0_3r1_x3i1, _MM_SHUFFLE(2, 3, 0, 1));
+    const __m128  x3_swapped = _mm_mul_ps(mm_swap_sign, x3i0_3r0_3i1_x3r1);
+    const __m128  x1_x3_add = _mm_add_ps(x1r0_1i0_1r1_x1i1, x3_swapped);
+    const __m128  x1_x3_sub = _mm_sub_ps(x1r0_1i0_1r1_x1i1, x3_swapped);
+    _mm_storel_epi64((__m128i*)&a[j0 +  8], (__m128i)x1_x3_add);
+    _mm_storel_epi64((__m128i*)&a[j0 + 24], (__m128i)x1_x3_sub);
+
+    const __m128 yy0 = _mm_shuffle_ps(x1_x3_add, x1_x3_sub,
+                                      _MM_SHUFFLE(2, 2, 2 ,2));
+    const __m128 yy1 = _mm_shuffle_ps(x1_x3_add, x1_x3_sub,
+                                      _MM_SHUFFLE(3, 3, 3 ,3));
+    const __m128 yy2 = _mm_mul_ps(mm_swap_sign, yy1);
+    const __m128 yy3 = _mm_add_ps(yy0, yy2);
+    const __m128 yy4 = _mm_mul_ps(wk1rv, yy3);
+    _mm_storel_epi64((__m128i*)&a[j0 + 40], (__m128i)yy4);
+    _mm_storel_epi64((__m128i*)&a[j0 + 56],
+                     _mm_shuffle_epi32((__m128i)yy4, _MM_SHUFFLE(2, 3, 2, 3)));
+  }
+
+  k1 = 0;
+  k = 64;
+  k1 += 2;
+  k2 = 2 * k1;
+  const __m128 wk2rv = _mm_load_ps(&rdft_wk2r[k2+0]);
+  const __m128 wk2iv = _mm_load_ps(&rdft_wk2i[k2+0]);
+               wk1rv = _mm_load_ps(&rdft_wk1r[k2+0]);
+  const __m128 wk1iv = _mm_load_ps(&rdft_wk1i[k2+0]);
+  const __m128 wk3rv = _mm_load_ps(&rdft_wk3r[k2+0]);
+  const __m128 wk3iv = _mm_load_ps(&rdft_wk3i[k2+0]);
+  for (j0 = k; j0 < l + k; j0 += 2) {
+    const __m128i a_00 = _mm_loadl_epi64((__m128i*)&a[j0 +  0]);
+    const __m128i a_08 = _mm_loadl_epi64((__m128i*)&a[j0 +  8]);
+    const __m128i a_32 = _mm_loadl_epi64((__m128i*)&a[j0 + 32]);
+    const __m128i a_40 = _mm_loadl_epi64((__m128i*)&a[j0 + 40]);
+    const __m128 a_00_32 = _mm_shuffle_ps((__m128)a_00, (__m128)a_32,
+                                          _MM_SHUFFLE(1, 0, 1 ,0));
+    const __m128 a_08_40 = _mm_shuffle_ps((__m128)a_08, (__m128)a_40,
+                                          _MM_SHUFFLE(1, 0, 1 ,0));
+          __m128 x0r0_0i0_0r1_x0i1 = _mm_add_ps(a_00_32, a_08_40);
+    const __m128 x1r0_1i0_1r1_x1i1 = _mm_sub_ps(a_00_32, a_08_40);
+
+    const __m128i a_16 = _mm_loadl_epi64((__m128i*)&a[j0 + 16]);
+    const __m128i a_24 = _mm_loadl_epi64((__m128i*)&a[j0 + 24]);
+    const __m128i a_48 = _mm_loadl_epi64((__m128i*)&a[j0 + 48]);
+    const __m128i a_56 = _mm_loadl_epi64((__m128i*)&a[j0 + 56]);
+    const __m128 a_16_48 = _mm_shuffle_ps((__m128)a_16, (__m128)a_48,
+                                          _MM_SHUFFLE(1, 0, 1 ,0));
+    const __m128 a_24_56 = _mm_shuffle_ps((__m128)a_24, (__m128)a_56,
+                                          _MM_SHUFFLE(1, 0, 1 ,0));
+    const __m128 x2r0_2i0_2r1_x2i1 = _mm_add_ps(a_16_48, a_24_56);
+    const __m128 x3r0_3i0_3r1_x3i1 = _mm_sub_ps(a_16_48, a_24_56);
+
+    const __m128 xx = _mm_add_ps(x0r0_0i0_0r1_x0i1, x2r0_2i0_2r1_x2i1);
+    _mm_storel_epi64((__m128i*)&a[j0 +  0], (__m128i)xx);
+    _mm_storel_epi64((__m128i*)&a[j0 + 32],
+                       _mm_shuffle_epi32((__m128i)xx, _MM_SHUFFLE(3, 2, 3, 2)));
+
+    const __m128 xx1 = _mm_sub_ps(x0r0_0i0_0r1_x0i1, x2r0_2i0_2r1_x2i1);
+    const __m128 xx2 = _mm_mul_ps(xx1 , wk2rv);
+    const __m128 xx3 = _mm_mul_ps(wk2iv,
+              (__m128)_mm_shuffle_epi32((__m128i)xx1, _MM_SHUFFLE(2, 3, 0, 1)));
+    const __m128 xx4 = _mm_add_ps(xx2, xx3);
+    _mm_storel_epi64((__m128i*)&a[j0 + 16], (__m128i)xx4);
+    _mm_storel_epi64((__m128i*)&a[j0 + 48],
+                      _mm_shuffle_epi32((__m128i)xx4, _MM_SHUFFLE(3, 2, 3, 2)));
+
+    const __m128  x3i0_3r0_3i1_x3r1 = (__m128)
+         _mm_shuffle_epi32((__m128i)x3r0_3i0_3r1_x3i1, _MM_SHUFFLE(2, 3, 0, 1));
+    const __m128  x3_swapped = _mm_mul_ps(mm_swap_sign, x3i0_3r0_3i1_x3r1);
+    const __m128  x1_x3_add = _mm_add_ps(x1r0_1i0_1r1_x1i1, x3_swapped);
+    const __m128  x1_x3_sub = _mm_sub_ps(x1r0_1i0_1r1_x1i1, x3_swapped);
+
+    const __m128 xx10 = _mm_mul_ps(x1_x3_add, wk1rv);
+    const __m128 xx11 = _mm_mul_ps(wk1iv,
+        (__m128)_mm_shuffle_epi32((__m128i)x1_x3_add, _MM_SHUFFLE(2, 3, 0, 1)));
+    const __m128 xx12 = _mm_add_ps(xx10, xx11);
+    _mm_storel_epi64((__m128i*)&a[j0 +  8], (__m128i)xx12);
+    _mm_storel_epi64((__m128i*)&a[j0 + 40],
+                     _mm_shuffle_epi32((__m128i)xx12, _MM_SHUFFLE(3, 2, 3, 2)));
+
+    const __m128 xx20 = _mm_mul_ps(x1_x3_sub, wk3rv);
+    const __m128 xx21 = _mm_mul_ps(wk3iv,
+        (__m128)_mm_shuffle_epi32((__m128i)x1_x3_sub, _MM_SHUFFLE(2, 3, 0, 1)));
+    const __m128 xx22 = _mm_add_ps(xx20, xx21);
+    _mm_storel_epi64((__m128i*)&a[j0 + 24], (__m128i)xx22);
+    _mm_storel_epi64((__m128i*)&a[j0 + 56],
+                     _mm_shuffle_epi32((__m128i)xx22, _MM_SHUFFLE(3, 2, 3, 2)));
+  }
+}
+
+static void rftfsub_128_SSE2(float *a) {
+  const float *c = rdft_w + 32;
+  int j1, j2, k1, k2;
+  float wkr, wki, xr, xi, yr, yi;
+
+  static const ALIGN16_BEG float ALIGN16_END k_half[4] =
+    {0.5f, 0.5f, 0.5f, 0.5f};
+  const __m128 mm_half = _mm_load_ps(k_half);
+
+  // Vectorized code (four at once).
+  //    Note: commented number are indexes for the first iteration of the loop.
+  for (j1 = 1, j2 = 2; j2 + 7 < 64; j1 += 4, j2 += 8) {
+    // Load 'wk'.
+    const __m128 c_j1 = _mm_loadu_ps(&c[     j1]);         //  1,  2,  3,  4,
+    const __m128 c_k1 = _mm_loadu_ps(&c[29 - j1]);         // 28, 29, 30, 31,
+    const __m128 wkrt = _mm_sub_ps(mm_half, c_k1);         // 28, 29, 30, 31,
+    const __m128 wkr_ =
+      _mm_shuffle_ps(wkrt, wkrt, _MM_SHUFFLE(0, 1, 2, 3)); // 31, 30, 29, 28,
+    const __m128 wki_ = c_j1;                              //  1,  2,  3,  4,
+    // Load and shuffle 'a'.
+    const __m128 a_j2_0 = _mm_loadu_ps(&a[0   + j2]);  //   2,   3,   4,   5,
+    const __m128 a_j2_4 = _mm_loadu_ps(&a[4   + j2]);  //   6,   7,   8,   9,
+    const __m128 a_k2_0 = _mm_loadu_ps(&a[122 - j2]);  // 120, 121, 122, 123,
+    const __m128 a_k2_4 = _mm_loadu_ps(&a[126 - j2]);  // 124, 125, 126, 127,
+    const __m128 a_j2_p0 = _mm_shuffle_ps(a_j2_0, a_j2_4,
+                            _MM_SHUFFLE(2, 0, 2 ,0));  //   2,   4,   6,   8,
+    const __m128 a_j2_p1 = _mm_shuffle_ps(a_j2_0, a_j2_4,
+                            _MM_SHUFFLE(3, 1, 3 ,1));  //   3,   5,   7,   9,
+    const __m128 a_k2_p0 = _mm_shuffle_ps(a_k2_4, a_k2_0,
+                            _MM_SHUFFLE(0, 2, 0 ,2));  // 126, 124, 122, 120,
+    const __m128 a_k2_p1 = _mm_shuffle_ps(a_k2_4, a_k2_0,
+                            _MM_SHUFFLE(1, 3, 1 ,3));  // 127, 125, 123, 121,
+    // Calculate 'x'.
+    const __m128 xr_ = _mm_sub_ps(a_j2_p0, a_k2_p0);
+                                               // 2-126, 4-124, 6-122, 8-120,
+    const __m128 xi_ = _mm_add_ps(a_j2_p1, a_k2_p1);
+                                               // 3-127, 5-125, 7-123, 9-121,
+    // Calculate product into 'y'.
+    //    yr = wkr * xr - wki * xi;
+    //    yi = wkr * xi + wki * xr;
+    const __m128 a_ = _mm_mul_ps(wkr_, xr_);
+    const __m128 b_ = _mm_mul_ps(wki_, xi_);
+    const __m128 c_ = _mm_mul_ps(wkr_, xi_);
+    const __m128 d_ = _mm_mul_ps(wki_, xr_);
+    const __m128 yr_ = _mm_sub_ps(a_, b_);     // 2-126, 4-124, 6-122, 8-120,
+    const __m128 yi_ = _mm_add_ps(c_, d_);     // 3-127, 5-125, 7-123, 9-121,
+    // Update 'a'.
+    //    a[j2 + 0] -= yr;
+    //    a[j2 + 1] -= yi;
+    //    a[k2 + 0] += yr;
+    //    a[k2 + 1] -= yi;
+    const __m128 a_j2_p0n = _mm_sub_ps(a_j2_p0, yr_);  //   2,   4,   6,   8,
+    const __m128 a_j2_p1n = _mm_sub_ps(a_j2_p1, yi_);  //   3,   5,   7,   9,
+    const __m128 a_k2_p0n = _mm_add_ps(a_k2_p0, yr_);  // 126, 124, 122, 120,
+    const __m128 a_k2_p1n = _mm_sub_ps(a_k2_p1, yi_);  // 127, 125, 123, 121,
+    // Shuffle in right order and store.
+    const __m128 a_j2_0n = _mm_unpacklo_ps(a_j2_p0n, a_j2_p1n);
+                                                       //   2,   3,   4,   5,
+    const __m128 a_j2_4n = _mm_unpackhi_ps(a_j2_p0n, a_j2_p1n);
+                                                       //   6,   7,   8,   9,
+    const __m128 a_k2_0nt = _mm_unpackhi_ps(a_k2_p0n, a_k2_p1n);
+                                                       // 122, 123, 120, 121,
+    const __m128 a_k2_4nt = _mm_unpacklo_ps(a_k2_p0n, a_k2_p1n);
+                                                       // 126, 127, 124, 125,
+    const __m128 a_k2_0n = _mm_shuffle_ps(a_k2_0nt, a_k2_0nt,
+                            _MM_SHUFFLE(1, 0, 3 ,2));  // 120, 121, 122, 123,
+    const __m128 a_k2_4n = _mm_shuffle_ps(a_k2_4nt, a_k2_4nt,
+                            _MM_SHUFFLE(1, 0, 3 ,2));  // 124, 125, 126, 127,
+    _mm_storeu_ps(&a[0   + j2], a_j2_0n);
+    _mm_storeu_ps(&a[4   + j2], a_j2_4n);
+    _mm_storeu_ps(&a[122 - j2], a_k2_0n);
+    _mm_storeu_ps(&a[126 - j2], a_k2_4n);
+  }
+  // Scalar code for the remaining items.
+  for (; j2 < 64; j1 += 1, j2 += 2) {
+    k2 = 128 - j2;
+    k1 =  32 - j1;
+    wkr = 0.5f - c[k1];
+    wki = c[j1];
+    xr = a[j2 + 0] - a[k2 + 0];
+    xi = a[j2 + 1] + a[k2 + 1];
+    yr = wkr * xr - wki * xi;
+    yi = wkr * xi + wki * xr;
+    a[j2 + 0] -= yr;
+    a[j2 + 1] -= yi;
+    a[k2 + 0] += yr;
+    a[k2 + 1] -= yi;
+  }
+}
+
+static void rftbsub_128_SSE2(float *a) {
+  const float *c = rdft_w + 32;
+  int j1, j2, k1, k2;
+  float wkr, wki, xr, xi, yr, yi;
+
+  static const ALIGN16_BEG float ALIGN16_END k_half[4] =
+    {0.5f, 0.5f, 0.5f, 0.5f};
+  const __m128 mm_half = _mm_load_ps(k_half);
+
+  a[1] = -a[1];
+  // Vectorized code (four at once).
+  //    Note: commented number are indexes for the first iteration of the loop.
+  for (j1 = 1, j2 = 2; j2 + 7 < 64; j1 += 4, j2 += 8) {
+    // Load 'wk'.
+    const __m128 c_j1 = _mm_loadu_ps(&c[     j1]);         //  1,  2,  3,  4,
+    const __m128 c_k1 = _mm_loadu_ps(&c[29 - j1]);         // 28, 29, 30, 31,
+    const __m128 wkrt = _mm_sub_ps(mm_half, c_k1);         // 28, 29, 30, 31,
+    const __m128 wkr_ =
+      _mm_shuffle_ps(wkrt, wkrt, _MM_SHUFFLE(0, 1, 2, 3)); // 31, 30, 29, 28,
+    const __m128 wki_ = c_j1;                              //  1,  2,  3,  4,
+    // Load and shuffle 'a'.
+    const __m128 a_j2_0 = _mm_loadu_ps(&a[0   + j2]);  //   2,   3,   4,   5,
+    const __m128 a_j2_4 = _mm_loadu_ps(&a[4   + j2]);  //   6,   7,   8,   9,
+    const __m128 a_k2_0 = _mm_loadu_ps(&a[122 - j2]);  // 120, 121, 122, 123,
+    const __m128 a_k2_4 = _mm_loadu_ps(&a[126 - j2]);  // 124, 125, 126, 127,
+    const __m128 a_j2_p0 = _mm_shuffle_ps(a_j2_0, a_j2_4,
+                            _MM_SHUFFLE(2, 0, 2 ,0));  //   2,   4,   6,   8,
+    const __m128 a_j2_p1 = _mm_shuffle_ps(a_j2_0, a_j2_4,
+                            _MM_SHUFFLE(3, 1, 3 ,1));  //   3,   5,   7,   9,
+    const __m128 a_k2_p0 = _mm_shuffle_ps(a_k2_4, a_k2_0,
+                            _MM_SHUFFLE(0, 2, 0 ,2));  // 126, 124, 122, 120,
+    const __m128 a_k2_p1 = _mm_shuffle_ps(a_k2_4, a_k2_0,
+                            _MM_SHUFFLE(1, 3, 1 ,3));  // 127, 125, 123, 121,
+    // Calculate 'x'.
+    const __m128 xr_ = _mm_sub_ps(a_j2_p0, a_k2_p0);
+                                               // 2-126, 4-124, 6-122, 8-120,
+    const __m128 xi_ = _mm_add_ps(a_j2_p1, a_k2_p1);
+                                               // 3-127, 5-125, 7-123, 9-121,
+    // Calculate product into 'y'.
+    //    yr = wkr * xr + wki * xi;
+    //    yi = wkr * xi - wki * xr;
+    const __m128 a_ = _mm_mul_ps(wkr_, xr_);
+    const __m128 b_ = _mm_mul_ps(wki_, xi_);
+    const __m128 c_ = _mm_mul_ps(wkr_, xi_);
+    const __m128 d_ = _mm_mul_ps(wki_, xr_);
+    const __m128 yr_ = _mm_add_ps(a_, b_);     // 2-126, 4-124, 6-122, 8-120,
+    const __m128 yi_ = _mm_sub_ps(c_, d_);     // 3-127, 5-125, 7-123, 9-121,
+    // Update 'a'.
+    //    a[j2 + 0] = a[j2 + 0] - yr;
+    //    a[j2 + 1] = yi - a[j2 + 1];
+    //    a[k2 + 0] = yr + a[k2 + 0];
+    //    a[k2 + 1] = yi - a[k2 + 1];
+    const __m128 a_j2_p0n = _mm_sub_ps(a_j2_p0, yr_);  //   2,   4,   6,   8,
+    const __m128 a_j2_p1n = _mm_sub_ps(yi_, a_j2_p1);  //   3,   5,   7,   9,
+    const __m128 a_k2_p0n = _mm_add_ps(a_k2_p0, yr_);  // 126, 124, 122, 120,
+    const __m128 a_k2_p1n = _mm_sub_ps(yi_, a_k2_p1);  // 127, 125, 123, 121,
+    // Shuffle in right order and store.
+    const __m128 a_j2_0n = _mm_unpacklo_ps(a_j2_p0n, a_j2_p1n);
+                                                       //   2,   3,   4,   5,
+    const __m128 a_j2_4n = _mm_unpackhi_ps(a_j2_p0n, a_j2_p1n);
+                                                       //   6,   7,   8,   9,
+    const __m128 a_k2_0nt = _mm_unpackhi_ps(a_k2_p0n, a_k2_p1n);
+                                                       // 122, 123, 120, 121,
+    const __m128 a_k2_4nt = _mm_unpacklo_ps(a_k2_p0n, a_k2_p1n);
+                                                       // 126, 127, 124, 125,
+    const __m128 a_k2_0n = _mm_shuffle_ps(a_k2_0nt, a_k2_0nt,
+                            _MM_SHUFFLE(1, 0, 3 ,2));  // 120, 121, 122, 123,
+    const __m128 a_k2_4n = _mm_shuffle_ps(a_k2_4nt, a_k2_4nt,
+                            _MM_SHUFFLE(1, 0, 3 ,2));  // 124, 125, 126, 127,
+    _mm_storeu_ps(&a[0   + j2], a_j2_0n);
+    _mm_storeu_ps(&a[4   + j2], a_j2_4n);
+    _mm_storeu_ps(&a[122 - j2], a_k2_0n);
+    _mm_storeu_ps(&a[126 - j2], a_k2_4n);
+  }
+  // Scalar code for the remaining items.
+  for (; j2 < 64; j1 += 1, j2 += 2) {
+    k2 = 128 - j2;
+    k1 =  32 - j1;
+    wkr = 0.5f - c[k1];
+    wki = c[j1];
+    xr = a[j2 + 0] - a[k2 + 0];
+    xi = a[j2 + 1] + a[k2 + 1];
+    yr = wkr * xr + wki * xi;
+    yi = wkr * xi - wki * xr;
+    a[j2 + 0] = a[j2 + 0] - yr;
+    a[j2 + 1] = yi - a[j2 + 1];
+    a[k2 + 0] = yr + a[k2 + 0];
+    a[k2 + 1] = yi - a[k2 + 1];
+  }
+  a[65] = -a[65];
+}
+
+void aec_rdft_init_sse2(void) {
+  cft1st_128 = cft1st_128_SSE2;
+  cftmdl_128 = cftmdl_128_SSE2;
+  rftfsub_128 = rftfsub_128_SSE2;
+  rftbsub_128 = rftbsub_128_SSE2;
+}
+
+#endif  // WEBRTC_USE_SS2

src/modules/audio_processing/aec/main/source/echo_cancellation.c

@@ -0,0 +1,824 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+/*
+ * Contains the API functions for the AEC.
+ */
+#include <stdlib.h>
+#include <string.h>
+
+#include "echo_cancellation.h"
+#include "aec_core.h"
+#include "ring_buffer.h"
+#include "resampler.h"
+#ifdef AEC_DEBUG
+    #include <stdio.h>
+#endif
+
+#define BUF_SIZE_FRAMES 50 // buffer size (frames)
+// Maximum length of resampled signal. Must be an integer multiple of frames
+// (ceil(1/(1 + MIN_SKEW)*2) + 1)*FRAME_LEN
+// The factor of 2 handles wb, and the + 1 is as a safety margin
+#define MAX_RESAMP_LEN (5 * FRAME_LEN)
+
+static const int bufSizeSamp = BUF_SIZE_FRAMES * FRAME_LEN; // buffer size (samples)
+static const int sampMsNb = 8; // samples per ms in nb
+// Target suppression levels for nlp modes
+// log{0.001, 0.00001, 0.00000001}
+static const float targetSupp[3] = {-6.9f, -11.5f, -18.4f};
+static const float minOverDrive[3] = {1.0f, 2.0f, 5.0f};
+static const int initCheck = 42;
+
+typedef struct {
+    int delayCtr;
+    int sampFreq;
+    int splitSampFreq;
+    int scSampFreq;
+    float sampFactor; // scSampRate / sampFreq
+    short nlpMode;
+    short autoOnOff;
+    short activity;
+    short skewMode;
+    short bufSizeStart;
+    //short bufResetCtr;  // counts number of noncausal frames
+    int knownDelay;
+
+    // Stores the last frame added to the farend buffer
+    short farendOld[2][FRAME_LEN];
+    short initFlag; // indicates if AEC has been initialized
+
+    // Variables used for averaging far end buffer size
+    short counter;
+    short sum;
+    short firstVal;
+    short checkBufSizeCtr;
+
+    // Variables used for delay shifts
+    short msInSndCardBuf;
+    short filtDelay;
+    int timeForDelayChange;
+    int ECstartup;
+    int checkBuffSize;
+    int delayChange;
+    short lastDelayDiff;
+
+#ifdef AEC_DEBUG
+    FILE *bufFile;
+    FILE *delayFile;
+    FILE *skewFile;
+    FILE *preCompFile;
+    FILE *postCompFile;
+#endif // AEC_DEBUG
+
+    // Structures
+    void *farendBuf;
+    void *resampler;
+
+    int skewFrCtr;
+    int resample; // if the skew is small enough we don't resample
+    int highSkewCtr;
+    float skew;
+
+    int lastError;
+
+    aec_t *aec;
+} aecpc_t;
+
+// Estimates delay to set the position of the farend buffer read pointer
+// (controlled by knownDelay)
+static int EstBufDelay(aecpc_t *aecInst, short msInSndCardBuf);
+
+// Stuffs the farend buffer if the estimated delay is too large
+static int DelayComp(aecpc_t *aecInst);
+
+WebRtc_Word32 WebRtcAec_Create(void **aecInst)
+{
+    aecpc_t *aecpc;
+    if (aecInst == NULL) {
+        return -1;
+    }
+
+    aecpc = malloc(sizeof(aecpc_t));
+    *aecInst = aecpc;
+    if (aecpc == NULL) {
+        return -1;
+    }
+
+    if (WebRtcAec_CreateAec(&aecpc->aec) == -1) {
+        WebRtcAec_Free(aecpc);
+        aecpc = NULL;
+        return -1;
+    }
+
+    if (WebRtcApm_CreateBuffer(&aecpc->farendBuf, bufSizeSamp) == -1) {
+        WebRtcAec_Free(aecpc);
+        aecpc = NULL;
+        return -1;
+    }
+
+    if (WebRtcAec_CreateResampler(&aecpc->resampler) == -1) {
+        WebRtcAec_Free(aecpc);
+        aecpc = NULL;
+        return -1;
+    }
+
+    aecpc->initFlag = 0;
+    aecpc->lastError = 0;
+
+#ifdef AEC_DEBUG
+    aecpc->aec->farFile = fopen("aecFar.pcm","wb");
+    aecpc->aec->nearFile = fopen("aecNear.pcm","wb");
+    aecpc->aec->outFile = fopen("aecOut.pcm","wb");
+    aecpc->aec->outLpFile = fopen("aecOutLp.pcm","wb");
+
+    aecpc->bufFile = fopen("aecBuf.dat", "wb");
+    aecpc->skewFile = fopen("aecSkew.dat", "wb");
+    aecpc->delayFile = fopen("aecDelay.dat", "wb");
+    aecpc->preCompFile = fopen("preComp.pcm", "wb");
+    aecpc->postCompFile = fopen("postComp.pcm", "wb");
+#endif // AEC_DEBUG
+
+    return 0;
+}
+
+WebRtc_Word32 WebRtcAec_Free(void *aecInst)
+{
+    aecpc_t *aecpc = aecInst;
+
+    if (aecpc == NULL) {
+        return -1;
+    }
+
+#ifdef AEC_DEBUG
+    fclose(aecpc->aec->farFile);
+    fclose(aecpc->aec->nearFile);
+    fclose(aecpc->aec->outFile);
+    fclose(aecpc->aec->outLpFile);
+
+    fclose(aecpc->bufFile);
+    fclose(aecpc->skewFile);
+    fclose(aecpc->delayFile);
+    fclose(aecpc->preCompFile);
+    fclose(aecpc->postCompFile);
+#endif // AEC_DEBUG
+
+    WebRtcAec_FreeAec(aecpc->aec);
+    WebRtcApm_FreeBuffer(aecpc->farendBuf);
+    WebRtcAec_FreeResampler(aecpc->resampler);
+    free(aecpc);
+
+    return 0;
+}
+
+WebRtc_Word32 WebRtcAec_Init(void *aecInst, WebRtc_Word32 sampFreq, WebRtc_Word32 scSampFreq)
+{
+    aecpc_t *aecpc = aecInst;
+    AecConfig aecConfig;
+
+    if (aecpc == NULL) {
+        return -1;
+    }
+
+    if (sampFreq != 8000 && sampFreq != 16000  && sampFreq != 32000) {
+        aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
+        return -1;
+    }
+    aecpc->sampFreq = sampFreq;
+
+    if (scSampFreq < 1 || scSampFreq > 96000) {
+        aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
+        return -1;
+    }
+    aecpc->scSampFreq = scSampFreq;
+
+    // Initialize echo canceller core
+    if (WebRtcAec_InitAec(aecpc->aec, aecpc->sampFreq) == -1) {
+        aecpc->lastError = AEC_UNSPECIFIED_ERROR;
+        return -1;
+    }
+
+    // Initialize farend buffer
+    if (WebRtcApm_InitBuffer(aecpc->farendBuf) == -1) {
+        aecpc->lastError = AEC_UNSPECIFIED_ERROR;
+        return -1;
+    }
+
+    if (WebRtcAec_InitResampler(aecpc->resampler, aecpc->scSampFreq) == -1) {
+        aecpc->lastError = AEC_UNSPECIFIED_ERROR;
+        return -1;
+    }
+
+    aecpc->initFlag = initCheck;  // indicates that initilisation has been done
+
+    if (aecpc->sampFreq == 32000) {
+        aecpc->splitSampFreq = 16000;
+    }
+    else {
+        aecpc->splitSampFreq = sampFreq;
+    }
+
+    aecpc->skewFrCtr = 0;
+    aecpc->activity = 0;
+
+    aecpc->delayChange = 1;
+    aecpc->delayCtr = 0;
+
+    aecpc->sum = 0;
+    aecpc->counter = 0;
+    aecpc->checkBuffSize = 1;
+    aecpc->firstVal = 0;
+
+    aecpc->ECstartup = 1;
+    aecpc->bufSizeStart = 0;
+    aecpc->checkBufSizeCtr = 0;
+    aecpc->filtDelay = 0;
+    aecpc->timeForDelayChange =0;
+    aecpc->knownDelay = 0;
+    aecpc->lastDelayDiff = 0;
+
+    aecpc->skew = 0;
+    aecpc->resample = kAecFalse;
+    aecpc->highSkewCtr = 0;
+    aecpc->sampFactor = (aecpc->scSampFreq * 1.0f) / aecpc->splitSampFreq;
+
+    memset(&aecpc->farendOld[0][0], 0, 160);
+
+    // Default settings.
+    aecConfig.nlpMode = kAecNlpModerate;
+    aecConfig.skewMode = kAecFalse;
+    aecConfig.metricsMode = kAecFalse;
+
+    if (WebRtcAec_set_config(aecpc, aecConfig) == -1) {
+        aecpc->lastError = AEC_UNSPECIFIED_ERROR;
+        return -1;
+    }
+
+    return 0;
+}
+
+// only buffer L band for farend
+WebRtc_Word32 WebRtcAec_BufferFarend(void *aecInst, const WebRtc_Word16 *farend,
+    WebRtc_Word16 nrOfSamples)
+{
+    aecpc_t *aecpc = aecInst;
+    WebRtc_Word32 retVal = 0;
+    short newNrOfSamples;
+    short newFarend[MAX_RESAMP_LEN];
+    float skew;
+
+    if (aecpc == NULL) {
+        return -1;
+    }
+
+    if (farend == NULL) {
+        aecpc->lastError = AEC_NULL_POINTER_ERROR;
+        return -1;
+    }
+
+    if (aecpc->initFlag != initCheck) {
+        aecpc->lastError = AEC_UNINITIALIZED_ERROR;
+        return -1;
+    }
+
+    // number of samples == 160 for SWB input
+    if (nrOfSamples != 80 && nrOfSamples != 160) {
+        aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
+        return -1;
+    }
+
+    skew = aecpc->skew;
+
+    // TODO: Is this really a good idea?
+    if (!aecpc->ECstartup) {
+        DelayComp(aecpc);
+    }
+
+    if (aecpc->skewMode == kAecTrue && aecpc->resample == kAecTrue) {
+        // Resample and get a new number of samples
+        newNrOfSamples = WebRtcAec_ResampleLinear(aecpc->resampler,
+                                                  farend,
+                                                  nrOfSamples,
+                                                  skew,
+                                                  newFarend);
+        WebRtcApm_WriteBuffer(aecpc->farendBuf, newFarend, newNrOfSamples);
+
+#ifdef AEC_DEBUG
+        fwrite(farend, 2, nrOfSamples, aecpc->preCompFile);
+        fwrite(newFarend, 2, newNrOfSamples, aecpc->postCompFile);
+#endif
+    }
+    else {
+        WebRtcApm_WriteBuffer(aecpc->farendBuf, farend, nrOfSamples);
+    }
+
+    return retVal;
+}
+
+WebRtc_Word32 WebRtcAec_Process(void *aecInst, const WebRtc_Word16 *nearend,
+    const WebRtc_Word16 *nearendH, WebRtc_Word16 *out, WebRtc_Word16 *outH,
+    WebRtc_Word16 nrOfSamples, WebRtc_Word16 msInSndCardBuf, WebRtc_Word32 skew)
+{
+    aecpc_t *aecpc = aecInst;
+    WebRtc_Word32 retVal = 0;
+    short i;
+    short farend[FRAME_LEN];
+    short nmbrOfFilledBuffers;
+    short nBlocks10ms;
+    short nFrames;
+#ifdef AEC_DEBUG
+    short msInAECBuf;
+#endif
+    // Limit resampling to doubling/halving of signal
+    const float minSkewEst = -0.5f;
+    const float maxSkewEst = 1.0f;
+
+    if (aecpc == NULL) {
+        return -1;
+    }
+
+    if (nearend == NULL) {
+        aecpc->lastError = AEC_NULL_POINTER_ERROR;
+        return -1;
+    }
+
+    if (out == NULL) {
+        aecpc->lastError = AEC_NULL_POINTER_ERROR;
+        return -1;
+    }
+
+    if (aecpc->initFlag != initCheck) {
+        aecpc->lastError = AEC_UNINITIALIZED_ERROR;
+        return -1;
+    }
+
+    // number of samples == 160 for SWB input
+    if (nrOfSamples != 80 && nrOfSamples != 160) {
+        aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
+        return -1;
+    }
+
+    // Check for valid pointers based on sampling rate
+    if (aecpc->sampFreq == 32000 && nearendH == NULL) {
+       aecpc->lastError = AEC_NULL_POINTER_ERROR;
+       return -1;
+    }
+
+    if (msInSndCardBuf < 0) {
+        msInSndCardBuf = 0;
+        aecpc->lastError = AEC_BAD_PARAMETER_WARNING;
+        retVal = -1;
+    }
+    else if (msInSndCardBuf > 500) {
+        msInSndCardBuf = 500;
+        aecpc->lastError = AEC_BAD_PARAMETER_WARNING;
+        retVal = -1;
+    }
+    msInSndCardBuf += 10;
+    aecpc->msInSndCardBuf = msInSndCardBuf;
+
+    if (aecpc->skewMode == kAecTrue) {
+        if (aecpc->skewFrCtr < 25) {
+            aecpc->skewFrCtr++;
+        }
+        else {
+            retVal = WebRtcAec_GetSkew(aecpc->resampler, skew, &aecpc->skew);
+            if (retVal == -1) {
+                aecpc->skew = 0;
+                aecpc->lastError = AEC_BAD_PARAMETER_WARNING;
+            }
+
+            aecpc->skew /= aecpc->sampFactor*nrOfSamples;
+
+            if (aecpc->skew < 1.0e-3 && aecpc->skew > -1.0e-3) {
+                aecpc->resample = kAecFalse;
+            }
+            else {
+                aecpc->resample = kAecTrue;
+            }
+
+            if (aecpc->skew < minSkewEst) {
+                aecpc->skew = minSkewEst;
+            }
+            else if (aecpc->skew > maxSkewEst) {
+                aecpc->skew = maxSkewEst;
+            }
+
+#ifdef AEC_DEBUG
+            fwrite(&aecpc->skew, sizeof(aecpc->skew), 1, aecpc->skewFile);
+#endif
+        }
+    }
+
+    nFrames = nrOfSamples / FRAME_LEN;
+    nBlocks10ms = nFrames / aecpc->aec->mult;
+
+    if (aecpc->ECstartup) {
+        if (nearend != out) {
+            // Only needed if they don't already point to the same place.
+            memcpy(out, nearend, sizeof(short) * nrOfSamples);
+        }
+        nmbrOfFilledBuffers = WebRtcApm_get_buffer_size(aecpc->farendBuf) / FRAME_LEN;
+
+        // The AEC is in the start up mode
+        // AEC is disabled until the soundcard buffer and farend buffers are OK
+
+        // Mechanism to ensure that the soundcard buffer is reasonably stable.
+        if (aecpc->checkBuffSize) {
+
+            aecpc->checkBufSizeCtr++;
+            // Before we fill up the far end buffer we require the amount of data on the
+            // sound card to be stable (+/-8 ms) compared to the first value. This
+            // comparison is made during the following 4 consecutive frames. If it seems
+            // to be stable then we start to fill up the far end buffer.
+
+            if (aecpc->counter == 0) {
+                aecpc->firstVal = aecpc->msInSndCardBuf;
+                aecpc->sum = 0;
+            }
+
+            if (abs(aecpc->firstVal - aecpc->msInSndCardBuf) <
+                WEBRTC_SPL_MAX(0.2 * aecpc->msInSndCardBuf, sampMsNb)) {
+                aecpc->sum += aecpc->msInSndCardBuf;
+                aecpc->counter++;
+            }
+            else {
+                aecpc->counter = 0;
+            }
+
+            if (aecpc->counter*nBlocks10ms >= 6) {
+                // The farend buffer size is determined in blocks of 80 samples
+                // Use 75% of the average value of the soundcard buffer
+                aecpc->bufSizeStart = WEBRTC_SPL_MIN((int) (0.75 * (aecpc->sum *
+                    aecpc->aec->mult) / (aecpc->counter * 10)), BUF_SIZE_FRAMES);
+                // buffersize has now been determined
+                aecpc->checkBuffSize = 0;
+            }
+
+            if (aecpc->checkBufSizeCtr * nBlocks10ms > 50) {
+                // for really bad sound cards, don't disable echocanceller for more than 0.5 sec
+                aecpc->bufSizeStart = WEBRTC_SPL_MIN((int) (0.75 * (aecpc->msInSndCardBuf *
+                    aecpc->aec->mult) / 10), BUF_SIZE_FRAMES);
+                aecpc->checkBuffSize = 0;
+            }
+        }
+
+        // if checkBuffSize changed in the if-statement above
+        if (!aecpc->checkBuffSize) {
+            // soundcard buffer is now reasonably stable
+            // When the far end buffer is filled with approximately the same amount of
+            // data as the amount on the sound card we end the start up phase and start
+            // to cancel echoes.
+
+            if (nmbrOfFilledBuffers == aecpc->bufSizeStart) {
+                aecpc->ECstartup = 0;  // Enable the AEC
+            }
+            else if (nmbrOfFilledBuffers > aecpc->bufSizeStart) {
+                WebRtcApm_FlushBuffer(aecpc->farendBuf, WebRtcApm_get_buffer_size(aecpc->farendBuf) -
+                    aecpc->bufSizeStart * FRAME_LEN);
+                aecpc->ECstartup = 0;
+            }
+        }
+
+    }
+    else {
+        // AEC is enabled
+
+        // Note only 1 block supported for nb and 2 blocks for wb
+        for (i = 0; i < nFrames; i++) {
+            nmbrOfFilledBuffers = WebRtcApm_get_buffer_size(aecpc->farendBuf) / FRAME_LEN;
+
+            // Check that there is data in the far end buffer
+            if (nmbrOfFilledBuffers > 0) {
+                // Get the next 80 samples from the farend buffer
+                WebRtcApm_ReadBuffer(aecpc->farendBuf, farend, FRAME_LEN);
+
+                // Always store the last frame for use when we run out of data
+                memcpy(&(aecpc->farendOld[i][0]), farend, FRAME_LEN * sizeof(short));
+            }
+            else {
+                // We have no data so we use the last played frame
+                memcpy(farend, &(aecpc->farendOld[i][0]), FRAME_LEN * sizeof(short));
+            }
+
+            // Call buffer delay estimator when all data is extracted,
+            // i.e. i = 0 for NB and i = 1 for WB or SWB
+            if ((i == 0 && aecpc->splitSampFreq == 8000) ||
+                    (i == 1 && (aecpc->splitSampFreq == 16000))) {
+                EstBufDelay(aecpc, aecpc->msInSndCardBuf);
+            }
+
+            // Call the AEC
+           WebRtcAec_ProcessFrame(aecpc->aec, farend, &nearend[FRAME_LEN * i], &nearendH[FRAME_LEN * i],
+               &out[FRAME_LEN * i], &outH[FRAME_LEN * i], aecpc->knownDelay);
+        }
+    }
+
+#ifdef AEC_DEBUG
+    msInAECBuf = WebRtcApm_get_buffer_size(aecpc->farendBuf) / (sampMsNb*aecpc->aec->mult);
+    fwrite(&msInAECBuf, 2, 1, aecpc->bufFile);
+    fwrite(&(aecpc->knownDelay), sizeof(aecpc->knownDelay), 1, aecpc->delayFile);
+#endif
+
+    return retVal;
+}
+
+WebRtc_Word32 WebRtcAec_set_config(void *aecInst, AecConfig config)
+{
+    aecpc_t *aecpc = aecInst;
+
+    if (aecpc == NULL) {
+        return -1;
+    }
+
+    if (aecpc->initFlag != initCheck) {
+        aecpc->lastError = AEC_UNINITIALIZED_ERROR;
+        return -1;
+    }
+
+    if (config.skewMode != kAecFalse && config.skewMode != kAecTrue) {
+        aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
+        return -1;
+    }
+    aecpc->skewMode = config.skewMode;
+
+    if (config.nlpMode != kAecNlpConservative && config.nlpMode !=
+            kAecNlpModerate && config.nlpMode != kAecNlpAggressive) {
+        aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
+        return -1;
+    }
+    aecpc->nlpMode = config.nlpMode;
+    aecpc->aec->targetSupp = targetSupp[aecpc->nlpMode];
+    aecpc->aec->minOverDrive = minOverDrive[aecpc->nlpMode];
+
+    if (config.metricsMode != kAecFalse && config.metricsMode != kAecTrue) {
+        aecpc->lastError = AEC_BAD_PARAMETER_ERROR;
+        return -1;
+    }
+    aecpc->aec->metricsMode = config.metricsMode;
+    if (aecpc->aec->metricsMode == kAecTrue) {
+        WebRtcAec_InitMetrics(aecpc->aec);
+    }
+
+    return 0;
+}
+
+WebRtc_Word32 WebRtcAec_get_config(void *aecInst, AecConfig *config)
+{
+    aecpc_t *aecpc = aecInst;
+
+    if (aecpc == NULL) {
+        return -1;
+    }
+
+    if (config == NULL) {
+        aecpc->lastError = AEC_NULL_POINTER_ERROR;
+        return -1;
+    }
+
+    if (aecpc->initFlag != initCheck) {
+        aecpc->lastError = AEC_UNINITIALIZED_ERROR;
+        return -1;
+    }
+
+    config->nlpMode = aecpc->nlpMode;
+    config->skewMode = aecpc->skewMode;
+    config->metricsMode = aecpc->aec->metricsMode;
+
+    return 0;
+}
+
+WebRtc_Word32 WebRtcAec_get_echo_status(void *aecInst, WebRtc_Word16 *status)
+{
+    aecpc_t *aecpc = aecInst;
+
+    if (aecpc == NULL) {
+        return -1;
+    }
+
+    if (status == NULL) {
+        aecpc->lastError = AEC_NULL_POINTER_ERROR;
+        return -1;
+    }
+
+    if (aecpc->initFlag != initCheck) {
+        aecpc->lastError = AEC_UNINITIALIZED_ERROR;
+        return -1;
+    }
+
+    *status = aecpc->aec->echoState;
+
+    return 0;
+}
+
+WebRtc_Word32 WebRtcAec_GetMetrics(void *aecInst, AecMetrics *metrics)
+{
+    const float upweight = 0.7f;
+    float dtmp;
+    short stmp;
+    aecpc_t *aecpc = aecInst;
+
+    if (aecpc == NULL) {
+        return -1;
+    }
+
+    if (metrics == NULL) {
+        aecpc->lastError = AEC_NULL_POINTER_ERROR;
+        return -1;
+    }
+
+    if (aecpc->initFlag != initCheck) {
+        aecpc->lastError = AEC_UNINITIALIZED_ERROR;
+        return -1;
+    }
+
+    // ERL
+    metrics->erl.instant = (short) aecpc->aec->erl.instant;
+
+    if ((aecpc->aec->erl.himean > offsetLevel) && (aecpc->aec->erl.average > offsetLevel)) {
+    // Use a mix between regular average and upper part average
+        dtmp = upweight * aecpc->aec->erl.himean + (1 - upweight) * aecpc->aec->erl.average;
+        metrics->erl.average = (short) dtmp;
+    }
+    else {
+        metrics->erl.average = offsetLevel;
+    }
+
+    metrics->erl.max = (short) aecpc->aec->erl.max;
+
+    if (aecpc->aec->erl.min < (offsetLevel * (-1))) {
+        metrics->erl.min = (short) aecpc->aec->erl.min;
+    }
+    else {
+        metrics->erl.min = offsetLevel;
+    }
+
+    // ERLE
+    metrics->erle.instant = (short) aecpc->aec->erle.instant;
+
+    if ((aecpc->aec->erle.himean > offsetLevel) && (aecpc->aec->erle.average > offsetLevel)) {
+        // Use a mix between regular average and upper part average
+        dtmp =  upweight * aecpc->aec->erle.himean + (1 - upweight) * aecpc->aec->erle.average;
+        metrics->erle.average = (short) dtmp;
+    }
+    else {
+        metrics->erle.average = offsetLevel;
+    }
+
+    metrics->erle.max = (short) aecpc->aec->erle.max;
+
+    if (aecpc->aec->erle.min < (offsetLevel * (-1))) {
+        metrics->erle.min = (short) aecpc->aec->erle.min;
+    } else {
+        metrics->erle.min = offsetLevel;
+    }
+
+    // RERL
+    if ((metrics->erl.average > offsetLevel) && (metrics->erle.average > offsetLevel)) {
+        stmp = metrics->erl.average + metrics->erle.average;
+    }
+    else {
+        stmp = offsetLevel;
+    }
+    metrics->rerl.average = stmp;
+
+    // No other statistics needed, but returned for completeness
+    metrics->rerl.instant = stmp;
+    metrics->rerl.max = stmp;
+    metrics->rerl.min = stmp;
+
+    // A_NLP
+    metrics->aNlp.instant = (short) aecpc->aec->aNlp.instant;
+
+    if ((aecpc->aec->aNlp.himean > offsetLevel) && (aecpc->aec->aNlp.average > offsetLevel)) {
+        // Use a mix between regular average and upper part average
+        dtmp =  upweight * aecpc->aec->aNlp.himean + (1 - upweight) * aecpc->aec->aNlp.average;
+        metrics->aNlp.average = (short) dtmp;
+    }
+    else {
+        metrics->aNlp.average = offsetLevel;
+    }
+
+    metrics->aNlp.max = (short) aecpc->aec->aNlp.max;
+
+    if (aecpc->aec->aNlp.min < (offsetLevel * (-1))) {
+        metrics->aNlp.min = (short) aecpc->aec->aNlp.min;
+    }
+    else {
+        metrics->aNlp.min = offsetLevel;
+    }
+
+    return 0;
+}
+
+WebRtc_Word32 WebRtcAec_get_version(WebRtc_Word8 *versionStr, WebRtc_Word16 len)
+{
+    const char version[] = "AEC 2.5.0";
+    const short versionLen = (short)strlen(version) + 1; // +1 for null-termination
+
+    if (versionStr == NULL) {
+        return -1;
+    }
+
+    if (versionLen > len) {
+        return -1;
+    }
+
+    strncpy(versionStr, version, versionLen);
+    return 0;
+}
+
+WebRtc_Word32 WebRtcAec_get_error_code(void *aecInst)
+{
+    aecpc_t *aecpc = aecInst;
+
+    if (aecpc == NULL) {
+        return -1;
+    }
+
+    return aecpc->lastError;
+}
+
+static int EstBufDelay(aecpc_t *aecpc, short msInSndCardBuf)
+{
+    short delayNew, nSampFar, nSampSndCard;
+    short diff;
+
+    nSampFar = WebRtcApm_get_buffer_size(aecpc->farendBuf);
+    nSampSndCard = msInSndCardBuf * sampMsNb * aecpc->aec->mult;
+
+    delayNew = nSampSndCard - nSampFar;
+
+    // Account for resampling frame delay
+    if (aecpc->skewMode == kAecTrue && aecpc->resample == kAecTrue) {
+        delayNew -= kResamplingDelay;
+    }
+
+    if (delayNew < FRAME_LEN) {
+        WebRtcApm_FlushBuffer(aecpc->farendBuf, FRAME_LEN);
+        delayNew += FRAME_LEN;
+    }
+
+    aecpc->filtDelay = WEBRTC_SPL_MAX(0, (short)(0.8*aecpc->filtDelay + 0.2*delayNew));
+
+    diff = aecpc->filtDelay - aecpc->knownDelay;
+    if (diff > 224) {
+        if (aecpc->lastDelayDiff < 96) {
+            aecpc->timeForDelayChange = 0;
+        }
+        else {
+            aecpc->timeForDelayChange++;
+        }
+    }
+    else if (diff < 96 && aecpc->knownDelay > 0) {
+        if (aecpc->lastDelayDiff > 224) {
+            aecpc->timeForDelayChange = 0;
+        }
+        else {
+            aecpc->timeForDelayChange++;
+        }
+    }
+    else {
+        aecpc->timeForDelayChange = 0;
+    }
+    aecpc->lastDelayDiff = diff;
+
+    if (aecpc->timeForDelayChange > 25) {
+        aecpc->knownDelay = WEBRTC_SPL_MAX((int)aecpc->filtDelay - 160, 0);
+    }
+    return 0;
+}
+
+static int DelayComp(aecpc_t *aecpc)
+{
+    int nSampFar, nSampSndCard, delayNew, nSampAdd;
+    const int maxStuffSamp = 10 * FRAME_LEN;
+
+    nSampFar = WebRtcApm_get_buffer_size(aecpc->farendBuf);
+    nSampSndCard = aecpc->msInSndCardBuf * sampMsNb * aecpc->aec->mult;
+    delayNew = nSampSndCard - nSampFar;
+
+    // Account for resampling frame delay
+    if (aecpc->skewMode == kAecTrue && aecpc->resample == kAecTrue) {
+        delayNew -= kResamplingDelay;
+    }
+
+    if (delayNew > FAR_BUF_LEN - FRAME_LEN*aecpc->aec->mult) {
+        // The difference of the buffersizes is larger than the maximum
+        // allowed known delay. Compensate by stuffing the buffer.
+        nSampAdd = (int)(WEBRTC_SPL_MAX((int)(0.5 * nSampSndCard - nSampFar),
+                    FRAME_LEN));
+        nSampAdd = WEBRTC_SPL_MIN(nSampAdd, maxStuffSamp);
+
+        WebRtcApm_StuffBuffer(aecpc->farendBuf, nSampAdd);
+        aecpc->delayChange = 1; // the delay needs to be updated
+    }
+
+    return 0;
+}

src/modules/audio_processing/aec/main/source/resampler.c

@@ -0,0 +1,233 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+/* Resamples a signal to an arbitrary rate. Used by the AEC to compensate for clock
+ * skew by resampling the farend signal.
+ */
+
+#include <assert.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+
+#include "resampler.h"
+#include "aec_core.h"
+
+enum { kFrameBufferSize = FRAME_LEN * 4 };
+enum { kEstimateLengthFrames = 400 };
+
+typedef struct {
+    short buffer[kFrameBufferSize];
+    float position;
+
+    int deviceSampleRateHz;
+    int skewData[kEstimateLengthFrames];
+    int skewDataIndex;
+    float skewEstimate;
+} resampler_t;
+
+static int EstimateSkew(const int* rawSkew,
+                        int size,
+                        int absLimit,
+                        float *skewEst);
+
+int WebRtcAec_CreateResampler(void **resampInst)
+{
+    resampler_t *obj = malloc(sizeof(resampler_t));
+    *resampInst = obj;
+    if (obj == NULL) {
+        return -1;
+    }
+
+    return 0;
+}
+
+int WebRtcAec_InitResampler(void *resampInst, int deviceSampleRateHz)
+{
+    resampler_t *obj = (resampler_t*) resampInst;
+    memset(obj->buffer, 0, sizeof(obj->buffer));
+    obj->position = 0.0;
+
+    obj->deviceSampleRateHz = deviceSampleRateHz;
+    memset(obj->skewData, 0, sizeof(obj->skewData));
+    obj->skewDataIndex = 0;
+    obj->skewEstimate = 0.0;
+
+    return 0;
+}
+
+int WebRtcAec_FreeResampler(void *resampInst)
+{
+    resampler_t *obj = (resampler_t*) resampInst;
+    free(obj);
+
+    return 0;
+}
+
+int WebRtcAec_ResampleLinear(void *resampInst,
+                             const short *inspeech,
+                             int size,
+                             float skew,
+                             short *outspeech)
+{
+    resampler_t *obj = (resampler_t*) resampInst;
+
+    short *y;
+    float be, tnew, interp;
+    int tn, outsize, mm;
+
+    if (size < 0 || size > 2 * FRAME_LEN) {
+        return -1;
+    }
+
+    // Add new frame data in lookahead
+    memcpy(&obj->buffer[FRAME_LEN + kResamplingDelay],
+           inspeech,
+           size * sizeof(short));
+
+    // Sample rate ratio
+    be = 1 + skew;
+
+    // Loop over input frame
+    mm = 0;
+    y = &obj->buffer[FRAME_LEN]; // Point at current frame
+
+    tnew = be * mm + obj->position;
+    tn = (int) tnew;
+
+    while (tn < size) {
+
+        // Interpolation
+        interp = y[tn] + (tnew - tn) * (y[tn+1] - y[tn]);
+
+        if (interp > 32767) {
+            interp = 32767;
+        }
+        else if (interp < -32768) {
+            interp = -32768;
+        }
+
+        outspeech[mm] = (short) interp;
+        mm++;
+
+        tnew = be * mm + obj->position;
+        tn = (int) tnew;
+    }
+
+    outsize = mm;
+    obj->position += outsize * be - size;
+
+    // Shift buffer
+    memmove(obj->buffer,
+            &obj->buffer[size],
+            (kFrameBufferSize - size) * sizeof(short));
+
+    return outsize;
+}
+
+int WebRtcAec_GetSkew(void *resampInst, int rawSkew, float *skewEst)
+{
+    resampler_t *obj = (resampler_t*)resampInst;
+    int err = 0;
+
+    if (obj->skewDataIndex < kEstimateLengthFrames) {
+        obj->skewData[obj->skewDataIndex] = rawSkew;
+        obj->skewDataIndex++;
+    }
+    else if (obj->skewDataIndex == kEstimateLengthFrames) {
+        err = EstimateSkew(obj->skewData,
+                           kEstimateLengthFrames,
+                           obj->deviceSampleRateHz,
+                           skewEst);
+        obj->skewEstimate = *skewEst;
+        obj->skewDataIndex++;
+    }
+    else {
+        *skewEst = obj->skewEstimate;
+    }
+
+    return err;
+}
+
+int EstimateSkew(const int* rawSkew,
+                 int size,
+                 int deviceSampleRateHz,
+                 float *skewEst)
+{
+    const int absLimitOuter = (int)(0.04f * deviceSampleRateHz);
+    const int absLimitInner = (int)(0.0025f * deviceSampleRateHz);
+    int i = 0;
+    int n = 0;
+    float rawAvg = 0;
+    float err = 0;
+    float rawAbsDev = 0;
+    int upperLimit = 0;
+    int lowerLimit = 0;
+    float cumSum = 0;
+    float x = 0;
+    float x2 = 0;
+    float y = 0;
+    float xy = 0;
+    float xAvg = 0;
+    float denom = 0;
+    float skew = 0;
+
+    *skewEst = 0; // Set in case of error below.
+    for (i = 0; i < size; i++) {
+      if ((rawSkew[i] < absLimitOuter && rawSkew[i] > -absLimitOuter)) {
+        n++;
+        rawAvg += rawSkew[i];
+      }
+    }
+
+    if (n == 0) {
+      return -1;
+    }
+    assert(n > 0);
+    rawAvg /= n;
+
+    for (i = 0; i < size; i++) {
+      if ((rawSkew[i] < absLimitOuter && rawSkew[i] > -absLimitOuter)) {
+        err = rawSkew[i] - rawAvg;
+        rawAbsDev += err >= 0 ? err : -err;
+      }
+    }
+    assert(n > 0);
+    rawAbsDev /= n;
+    upperLimit = (int)(rawAvg + 5 * rawAbsDev + 1); // +1 for ceiling.
+    lowerLimit = (int)(rawAvg - 5 * rawAbsDev - 1); // -1 for floor.
+
+    n = 0;
+    for (i = 0; i < size; i++) {
+        if ((rawSkew[i] < absLimitInner && rawSkew[i] > -absLimitInner) ||
+            (rawSkew[i] < upperLimit && rawSkew[i] > lowerLimit)) {
+            n++;
+            cumSum += rawSkew[i];
+            x += n;
+            x2 += n*n;
+            y += cumSum;
+            xy += n * cumSum;
+        }
+    }
+
+    if (n == 0) {
+      return -1;
+    }
+    assert(n > 0);
+    xAvg = x / n;
+    denom = x2 - xAvg*x;
+
+    if (denom != 0) {
+        skew = (xy - xAvg*y) / denom;
+    }
+
+    *skewEst = skew;
+    return 0;
+}

src/modules/audio_processing/aec/main/source/resampler.h

@@ -0,0 +1,32 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_RESAMPLER_H_
+#define WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_RESAMPLER_H_
+
+enum { kResamplingDelay = 1 };
+
+// Unless otherwise specified, functions return 0 on success and -1 on error
+int WebRtcAec_CreateResampler(void **resampInst);
+int WebRtcAec_InitResampler(void *resampInst, int deviceSampleRateHz);
+int WebRtcAec_FreeResampler(void *resampInst);
+
+// Estimates skew from raw measurement.
+int WebRtcAec_GetSkew(void *resampInst, int rawSkew, float *skewEst);
+
+// Resamples input using linear interpolation.
+// Returns size of resampled array.
+int WebRtcAec_ResampleLinear(void *resampInst,
+                             const short *inspeech,
+                             int size,
+                             float skew,
+                             short *outspeech);
+
+#endif // WEBRTC_MODULES_AUDIO_PROCESSING_AEC_MAIN_SOURCE_RESAMPLER_H_

src/modules/audio_processing/aecm/main/interface/echo_control_mobile.h

@@ -0,0 +1,250 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AECM_MAIN_INTERFACE_ECHO_CONTROL_MOBILE_H_
+#define WEBRTC_MODULES_AUDIO_PROCESSING_AECM_MAIN_INTERFACE_ECHO_CONTROL_MOBILE_H_
+
+#include "typedefs.h"
+
+enum {
+    AecmFalse = 0,
+    AecmTrue
+};
+
+// Errors
+#define AECM_UNSPECIFIED_ERROR           12000
+#define AECM_UNSUPPORTED_FUNCTION_ERROR  12001
+#define AECM_UNINITIALIZED_ERROR         12002
+#define AECM_NULL_POINTER_ERROR          12003
+#define AECM_BAD_PARAMETER_ERROR         12004
+
+// Warnings
+#define AECM_BAD_PARAMETER_WARNING       12100
+
+typedef struct {
+    WebRtc_Word16 cngMode;            // AECM_FALSE, AECM_TRUE (default)
+    WebRtc_Word16 echoMode;           // 0, 1, 2, 3 (default), 4
+} AecmConfig;
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/*
+ * Allocates the memory needed by the AECM. The memory needs to be
+ * initialized separately using the WebRtcAecm_Init() function.
+ *
+ * Inputs                           Description
+ * -------------------------------------------------------------------
+ * void **aecmInst                  Pointer to the AECM instance to be
+ *                                  created and initialized
+ *
+ * Outputs                          Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word32 return             0: OK
+ *                                 -1: error
+ */
+WebRtc_Word32 WebRtcAecm_Create(void **aecmInst);
+
+/*
+ * This function releases the memory allocated by WebRtcAecm_Create()
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void *aecmInst               Pointer to the AECM instance
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word32  return        0: OK
+ *                             -1: error
+ */
+WebRtc_Word32 WebRtcAecm_Free(void *aecmInst);
+
+/*
+ * Initializes an AECM instance.
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void           *aecmInst     Pointer to the AECM instance
+ * WebRtc_Word32  sampFreq      Sampling frequency of data
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word32  return        0: OK
+ *                             -1: error
+ */
+WebRtc_Word32 WebRtcAecm_Init(void* aecmInst,
+                              WebRtc_Word32 sampFreq);
+
+/*
+ * Inserts an 80 or 160 sample block of data into the farend buffer.
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void           *aecmInst     Pointer to the AECM instance
+ * WebRtc_Word16  *farend       In buffer containing one frame of
+ *                              farend signal
+ * WebRtc_Word16  nrOfSamples   Number of samples in farend buffer
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word32  return        0: OK
+ *                             -1: error
+ */
+WebRtc_Word32 WebRtcAecm_BufferFarend(void* aecmInst,
+                                      const WebRtc_Word16* farend,
+                                      WebRtc_Word16 nrOfSamples);
+
+/*
+ * Runs the AECM on an 80 or 160 sample blocks of data.
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void           *aecmInst      Pointer to the AECM instance
+ * WebRtc_Word16  *nearendNoisy  In buffer containing one frame of
+ *                               reference nearend+echo signal. If
+ *                               noise reduction is active, provide
+ *                               the noisy signal here.
+ * WebRtc_Word16  *nearendClean  In buffer containing one frame of
+ *                               nearend+echo signal. If noise
+ *                               reduction is active, provide the
+ *                               clean signal here. Otherwise pass a
+ *                               NULL pointer.
+ * WebRtc_Word16  nrOfSamples    Number of samples in nearend buffer
+ * WebRtc_Word16  msInSndCardBuf Delay estimate for sound card and
+ *                               system buffers
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word16  *out          Out buffer, one frame of processed nearend
+ * WebRtc_Word32  return        0: OK
+ *                             -1: error
+ */
+WebRtc_Word32 WebRtcAecm_Process(void* aecmInst,
+                                 const WebRtc_Word16* nearendNoisy,
+                                 const WebRtc_Word16* nearendClean,
+                                 WebRtc_Word16* out,
+                                 WebRtc_Word16 nrOfSamples,
+                                 WebRtc_Word16 msInSndCardBuf);
+
+/*
+ * This function enables the user to set certain parameters on-the-fly
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void     *aecmInst           Pointer to the AECM instance
+ * AecmConfig config            Config instance that contains all
+ *                              properties to be set
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word32  return        0: OK
+ *                             -1: error
+ */
+WebRtc_Word32 WebRtcAecm_set_config(void* aecmInst,
+                                    AecmConfig config);
+
+/*
+ * This function enables the user to set certain parameters on-the-fly
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void *aecmInst               Pointer to the AECM instance
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * AecmConfig  *config          Pointer to the config instance that
+ *                              all properties will be written to
+ * WebRtc_Word32  return        0: OK
+ *                             -1: error
+ */
+WebRtc_Word32 WebRtcAecm_get_config(void *aecmInst,
+                                    AecmConfig *config);
+
+/*
+ * This function enables the user to set the echo path on-the-fly.
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void*        aecmInst        Pointer to the AECM instance
+ * void*        echo_path       Pointer to the echo path to be set
+ * size_t       size_bytes      Size in bytes of the echo path
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word32  return        0: OK
+ *                             -1: error
+ */
+WebRtc_Word32 WebRtcAecm_InitEchoPath(void* aecmInst,
+                                      const void* echo_path,
+                                      size_t size_bytes);
+
+/*
+ * This function enables the user to get the currently used echo path
+ * on-the-fly
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void*        aecmInst        Pointer to the AECM instance
+ * void*        echo_path       Pointer to echo path
+ * size_t       size_bytes      Size in bytes of the echo path
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word32  return        0: OK
+ *                             -1: error
+ */
+WebRtc_Word32 WebRtcAecm_GetEchoPath(void* aecmInst,
+                                     void* echo_path,
+                                     size_t size_bytes);
+
+/*
+ * This function enables the user to get the echo path size in bytes
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * size_t       return           : size in bytes
+ */
+size_t WebRtcAecm_echo_path_size_bytes();
+
+/*
+ * Gets the last error code.
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * void         *aecmInst       Pointer to the AECM instance
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word32  return        11000-11100: error code
+ */
+WebRtc_Word32 WebRtcAecm_get_error_code(void *aecmInst);
+
+/*
+ * Gets a version string
+ *
+ * Inputs                       Description
+ * -------------------------------------------------------------------
+ * char           *versionStr   Pointer to a string array
+ * WebRtc_Word16  len           The maximum length of the string
+ *
+ * Outputs                      Description
+ * -------------------------------------------------------------------
+ * WebRtc_Word8   *versionStr   Pointer to a string array
+ * WebRtc_Word32  return        0: OK
+ *                             -1: error
+ */
+WebRtc_Word32 WebRtcAecm_get_version(WebRtc_Word8 *versionStr,
+                                     WebRtc_Word16 len);
+
+#ifdef __cplusplus
+}
+#endif
+#endif /* WEBRTC_MODULES_AUDIO_PROCESSING_AECM_MAIN_INTERFACE_ECHO_CONTROL_MOBILE_H_ */

src/modules/audio_processing/aecm/main/matlab/compsup.m

@@ -0,0 +1,447 @@
+function [emicrophone,aaa]=compsup(microphone,TheFarEnd,avtime,samplingfreq);
+% microphone = microphone signal
+% aaa = nonlinearity input variable
+% TheFarEnd = far end signal
+% avtime = interval to compute suppression from (seconds)
+% samplingfreq = sampling frequency
+
+%if(nargin==6)
+%    fprintf(1,'suppress has received a delay sequence\n');
+%end
+
+
+Ap500=[  1.00, -4.95, 9.801, -9.70299, 4.80298005, -0.9509900499];
+Bp500=[  0.662743088639636, -2.5841655608125, 3.77668102146288, -2.45182477425154, 0.596566274575251, 0.0];
+
+
+Ap200=[ 1.00, -4.875, 9.50625, -9.26859375, 4.518439453125, -0.881095693359375];
+Bp200=[ 0.862545460994275, -3.2832804496114, 4.67892032308828, -2.95798023879133, 0.699796870041299, 0.0];
+
+maxDelay=0.4; %[s]
+histLen=1; %[s]
+
+
+% CONSTANTS THAT YOU CAN EXPERIMENT WITH
+A_GAIN=10.0;	 	% for the suppress case
+oversampling = 2;	% must be power of 2; minimum is 2; 4 works
+% fine for support=64, but for support=128,
+% 8 gives better results.
+support=64; %512	% fft support (frequency resolution; at low
+% settings you can hear more distortion
+% (e.g. pitch that is left-over from far-end))
+% 128 works well, 64 is ok)
+
+lowlevel = mean(abs(microphone))*0.0001;
+
+G_ol = 0;  % Use overlapping sets of estimates
+
+% ECHO SUPPRESSION SPECIFIC PARAMETERS
+suppress_overdrive=1.0;   % overdrive factor for suppression 1.4 is good
+gamma_echo=1.0;           % same as suppress_overdrive but at different place
+de_echo_bound=0.0;
+mLim=10;                  % rank of matrix G
+%limBW = 1;		  % use bandwidth-limited response for G
+if mLim > (support/2+1)
+    error('mLim in suppress.m too large\n');
+end
+
+
+dynrange=1.0000e-004;
+
+% other, constants
+hsupport = support/2;
+hsupport1 = hsupport+1;
+factor =  2 / oversampling;
+updatel = support/oversampling;
+win=sqrt(designwindow(0,support));
+estLen = round(avtime * samplingfreq/updatel)
+
+runningfmean =0.0;
+
+mLim = floor(hsupport1/2);
+V = sqrt(2/hsupport1)*cos(pi/hsupport1*(repmat((0:hsupport1-1) + 0.5, mLim, 1).* ...
+    repmat((0:mLim-1)' + 0.5, 1, hsupport1)));
+
+fprintf(1,'updatel is %5.3f s\n', updatel/samplingfreq);
+
+
+
+bandfirst=8; bandlast=25;
+dosmooth=0;  % to get rid of wavy bin counts (can be worse or better)
+
+% compute some constants
+blockLen = support/oversampling;
+maxDelayb = floor(samplingfreq*maxDelay/updatel); % in blocks
+histLenb = floor(samplingfreq*histLen/updatel); % in blocks
+
+x0=TheFarEnd;
+y0=microphone;
+
+
+%input
+tlength=min([length(microphone),length(TheFarEnd)]);
+updateno=floor(tlength/updatel);
+tlength=updatel*updateno;
+updateno = updateno - oversampling + 1;
+
+TheFarEnd =TheFarEnd(1:tlength);
+microphone =microphone(1:tlength);
+
+TheFarEnd =[zeros(hsupport,1);TheFarEnd(1:tlength)];
+microphone =[zeros(hsupport,1);microphone(1:tlength)];
+
+
+% signal length
+n = min([floor(length(x0)/support)*support,floor(length(y0)/support)*support]);
+nb = n/blockLen - oversampling + 1; % in blocks
+
+% initialize space
+win = sqrt([0 ; hanning(support-1)]);
+sxAll2 = zeros(hsupport1,nb);
+syAll2 = zeros(hsupport1,nb);
+
+z500=zeros(5,maxDelayb+1);
+z200=zeros(5,hsupport1);
+
+bxspectrum=uint32(zeros(nb,1));
+bxhist=uint32(zeros(maxDelayb+1,1));
+byspectrum=uint32(zeros(nb,1));
+bcount=zeros(1+maxDelayb,nb);
+fcount=zeros(1+maxDelayb,nb);
+fout=zeros(1+maxDelayb,nb);
+delay=zeros(nb,1);
+tdelay=zeros(nb,1);
+nlgains=zeros(nb,1);
+
+% create space (mainly for debugging)
+emicrophone=zeros(tlength,1);
+femicrophone=complex(zeros(hsupport1,updateno));
+thefilter=zeros(hsupport1,updateno);
+thelimiter=ones(hsupport1,updateno);
+fTheFarEnd=complex(zeros(hsupport1,updateno));
+afTheFarEnd=zeros(hsupport1,updateno);
+fmicrophone=complex(zeros(hsupport1,updateno));
+afmicrophone=zeros(hsupport1,updateno);
+
+G = zeros(hsupport1, hsupport1);
+zerovec = zeros(hsupport1,1);
+zeromat = zeros(hsupport1);
+
+% Reset sums
+mmxs_a = zerovec;
+mmys_a = zerovec;
+s2xs_a = zerovec;
+s2ys_a = zerovec;
+Rxxs_a = zeromat;
+Ryxs_a = zeromat;
+count_a = 1;
+
+mmxs_b = zerovec;
+mmys_b = zerovec;
+s2xs_b = zerovec;
+s2ys_b = zerovec;
+Rxxs_b = zeromat;
+Ryxs_b = zeromat;
+count_b = 1;
+
+nog=0;
+
+aaa=zeros(size(TheFarEnd));
+
+% loop over signal blocks
+fprintf(1,'.. Suppression; averaging G over %5.1f seconds; file length %5.1f seconds ..\n',avtime, length(microphone)/samplingfreq);
+fprintf(1,'.. SUPPRESSING ONLY AFTER %5.1f SECONDS! ..\n',avtime);
+fprintf(1,'.. 20 seconds is good ..\n');
+hh = waitbar_j(0,'Please wait...');
+
+
+for i=1:updateno
+
+    sb = (i-1)*updatel + 1;
+    se=sb+support-1;
+    
+    % analysis FFTs
+    temp=fft(win .* TheFarEnd(sb:se));
+    fTheFarEnd(:,i)=temp(1:hsupport1);
+    xf=fTheFarEnd(:,i);
+    afTheFarEnd(:,i)= abs(fTheFarEnd(:,i));
+    
+    temp=win .* microphone(sb:se);
+    
+    temp=fft(win .* microphone(sb:se));
+    fmicrophone(:,i)=temp(1:hsupport1);
+    yf=fmicrophone(:,i);
+    afmicrophone(:,i)= abs(fmicrophone(:,i));
+
+    
+    ener_orig = afmicrophone(:,i)'*afmicrophone(:,i);
+    if( ener_orig == 0)
+        afmicrophone(:,i)=lowlevel*ones(size(afmicrophone(:,i)));
+    end
+    
+    
+    	% use log domain (showed improved performance)
+xxf= sqrt(real(xf.*conj(xf))+1e-20);
+yyf= sqrt(real(yf.*conj(yf))+1e-20);
+        sxAll2(:,i) = 20*log10(xxf);
+	syAll2(:,i) = 20*log10(yyf);
+
+       mD=min(i-1,maxDelayb);
+      xthreshold = sum(sxAll2(:,i-mD:i),2)/(maxDelayb+1);
+
+      [yout, z200] = filter(Bp200,Ap200,syAll2(:,i),z200,2);
+      yout=yout/(maxDelayb+1);
+      ythreshold = mean(syAll2(:,i-mD:i),2);
+      
+
+  bxspectrum(i)=getBspectrum(sxAll2(:,i),xthreshold,bandfirst,bandlast);
+  byspectrum(i)=getBspectrum(syAll2(:,i),yout,bandfirst,bandlast);
+
+  bxhist(end-mD:end)=bxspectrum(i-mD:i);
+  
+  bcount(:,i)=hisser2( ...
+     byspectrum(i),flipud(bxhist),bandfirst,bandlast);
+ 
+ 
+  [fout(:,i), z500] = filter(Bp500,Ap500,bcount(:,i),z500,2);
+  fcount(:,i)=sum(bcount(:,max(1,i-histLenb+1):i),2); % using the history range
+ fout(:,i)=round(fout(:,i)); 
+  [value,delay(i)]=min(fout(:,i),[],1);
+  tdelay(i)=(delay(i)-1)*support/(samplingfreq*oversampling);
+
+    % compensate
+
+    idel =  max(i - delay(i) + 1,1);
+    
+  
+    % echo suppression
+    
+    noisyspec = afmicrophone(:,i);
+    
+    % Estimate G using covariance matrices
+    
+    % Cumulative estimates    
+    xx = afTheFarEnd(:,idel);
+    yy = afmicrophone(:,i);
+    
+    % Means
+    mmxs_a = mmxs_a + xx;
+    mmys_a = mmys_a + yy;
+    if (G_ol)
+        mmxs_b = mmxs_b + xx;  
+        mmys_b = mmys_b + yy;
+        mmy = mean([mmys_a/count_a mmys_b/count_b],2);
+        mmx = mean([mmxs_a/count_a mmxs_b/count_b],2);
+    else
+        mmx = mmxs_a/count_a;   
+        mmy = mmys_a/count_a;   
+    end
+    count_a = count_a + 1;
+    count_b = count_b + 1;
+    
+    % Mean removal
+    xxm = xx - mmx;
+    yym = yy - mmy;
+    
+    % Variances
+    s2xs_a = s2xs_a +  xxm .* xxm;
+    s2ys_a = s2ys_a +  yym .* yym;
+    s2xs_b = s2xs_b +  xxm .* xxm;
+    s2ys_b = s2ys_b +  yym .* yym;
+    
+    % Correlation matrices  
+    Rxxs_a = Rxxs_a + xxm * xxm';
+    Ryxs_a = Ryxs_a + yym * xxm';
+    Rxxs_b = Rxxs_b + xxm * xxm';
+    Ryxs_b = Ryxs_b + yym * xxm';
+    
+    
+    % Gain matrix A
+    
+    if mod(i, estLen) == 0
+        
+        
+        % Cumulative based estimates
+        Rxxf = Rxxs_a / (estLen - 1);
+        Ryxf = Ryxs_a / (estLen - 1);
+        
+        % Variance normalization
+        s2x2 = s2xs_a / (estLen - 1);
+        s2x2 = sqrt(s2x2);
+       % Sx = diag(max(s2x2,dynrange*max(s2x2)));
+        Sx = diag(s2x2);
+        if (sum(s2x2) > 0)
+          iSx = inv(Sx);
+         else
+                 iSx= Sx + 0.01;
+         end
+             
+        s2y2 = s2ys_a / (estLen - 1);
+        s2y2 = sqrt(s2y2);
+       % Sy = diag(max(s2y2,dynrange*max(s2y2)));
+        Sy = diag(s2y2);
+        iSy = inv(Sy);        
+        rx = iSx * Rxxf * iSx;
+        ryx = iSy * Ryxf * iSx;
+        
+     
+        
+        dbd= 7; % Us less than the full matrix
+        
+        % k x m
+        % Bandlimited structure on G
+        LSEon = 0; % Default is using MMSE
+        if (LSEon)
+            ryx = ryx*rx;
+            rx = rx*rx;
+        end
+        p = dbd-1;
+        gaj = min(min(hsupport1,2*p+1),min([p+(1:hsupport1); hsupport1+p+1-(1:hsupport1)]));
+        cgaj = [0 cumsum(gaj)];
+        
+        G3 = zeros(hsupport1);
+        for kk=1:hsupport1
+            ki = max(0,kk-p-1);
+            if (sum(sum(rx(ki+1:ki+gaj(kk),ki+1:ki+gaj(kk))))>0)
+               G3(kk,ki+1:ki+gaj(kk)) = ryx(kk,ki+1:ki+gaj(kk))/rx(ki+1:ki+gaj(kk),ki+1:ki+gaj(kk));
+           else
+               G3(kk,ki+1:ki+gaj(kk)) = ryx(kk,ki+1:ki+gaj(kk));
+           end
+        end
+        % End Bandlimited structure
+        
+        G = G3;
+        G(abs(G)<0.01)=0;
+        G = suppress_overdrive * Sy * G * iSx;
+        
+        if 1
+            figure(32); mi=2;
+            surf(max(min(G,mi),-mi)); view(2)
+            title('Unscaled Masked Limited-bandwidth G');
+        end
+        pause(0.05);
+        
+        % Reset sums
+        mmxs_a = zerovec;
+        mmys_a = zerovec;
+        s2xs_a = zerovec;
+        s2ys_a = zerovec;
+        Rxxs_a = zeromat;
+        Ryxs_a = zeromat;
+        count_a = 1;
+        
+    end
+    
+    if (G_ol)    
+        % Gain matrix B
+        
+        if ((mod((i-estLen/2), estLen) == 0) & i>estLen)
+            
+            
+            % Cumulative based estimates
+            Rxxf = Rxxs_b / (estLen - 1);
+            Ryxf = Ryxs_b / (estLen - 1);
+            
+            % Variance normalization
+            s2x2 = s2xs_b / (estLen - 1);
+            s2x2 = sqrt(s2x2);
+            Sx = diag(max(s2x2,dynrange*max(s2x2)));
+            iSx = inv(Sx);
+            s2y2 = s2ys_b / (estLen - 1);
+            s2y2 = sqrt(s2y2);
+            Sy = diag(max(s2y2,dynrange*max(s2y2)));
+            iSy = inv(Sy);        
+            rx = iSx * Rxxf * iSx;
+            ryx = iSy * Ryxf * iSx;
+            
+            
+            % Bandlimited structure on G
+            LSEon = 0; % Default is using MMSE
+            if (LSEon)
+                ryx = ryx*rx;
+                rx = rx*rx;
+            end
+            p = dbd-1;
+            gaj = min(min(hsupport1,2*p+1),min([p+(1:hsupport1); hsupport1+p+1-(1:hsupport1)]));
+            cgaj = [0 cumsum(gaj)];
+            
+            G3 = zeros(hsupport1);
+            for kk=1:hsupport1
+                ki = max(0,kk-p-1);
+                G3(kk,ki+1:ki+gaj(kk)) = ryx(kk,ki+1:ki+gaj(kk))/rx(ki+1:ki+gaj(kk),ki+1:ki+gaj(kk));
+            end
+            % End Bandlimited structure
+            
+            G = G3;
+            G(abs(G)<0.01)=0;
+            G = suppress_overdrive * Sy * G * iSx;
+            
+            if 1
+                figure(32); mi=2;
+                surf(max(min(G,mi),-mi)); view(2)
+                title('Unscaled Masked Limited-bandwidth G');
+            end
+            pause(0.05);
+            
+            
+            % Reset sums
+            mmxs_b = zerovec;
+            mmys_b = zerovec;
+            s2xs_b = zerovec;
+            s2ys_b = zerovec;
+            Rxxs_b = zeromat;
+            Ryxs_b = zeromat;
+            count_b = 1;
+            
+        end
+        
+    end
+    
+    FECestimate2 = G*afTheFarEnd(:,idel);
+    
+    % compute Wiener filter and suppressor function
+    thefilter(:,i) = (noisyspec - gamma_echo*FECestimate2) ./ noisyspec;
+    ix0 = find(thefilter(:,i)<de_echo_bound);   % bounding trick 1
+    thefilter(ix0,i) = de_echo_bound;     % bounding trick 2
+    ix0 = find(thefilter(:,i)>1);   % bounding in reasonable range
+    thefilter(ix0,i) = 1;
+    
+    % NONLINEARITY
+    nl_alpha=0.8;    % memory; seems not very critical
+    nlSeverity=0.3;  % nonlinearity severity: 0 does nothing; 1 suppresses all
+    thefmean=mean(thefilter(8:16,i));
+    if (thefmean<1)
+        disp('');
+    end
+    runningfmean = nl_alpha*runningfmean + (1-nl_alpha)*thefmean;
+    aaa(sb+20+1:sb+20+updatel)=10000*runningfmean* ones(updatel,1); % debug
+    slope0=1.0/(1.0-nlSeverity); %
+    thegain = max(0.0,min(1.0,slope0*(runningfmean-nlSeverity)));
+    % END NONLINEARITY
+    thefilter(:,i) = thegain*thefilter(:,i);
+    
+    
+    % Wiener filtering
+    femicrophone(:,i) = fmicrophone(:,i) .* thefilter(:,i);
+    thelimiter(:,i) = (noisyspec - A_GAIN*FECestimate2) ./ noisyspec;
+    index = find(thelimiter(:,i)>1.0);
+    thelimiter(index,i) = 1.0;
+    index = find(thelimiter(:,i)<0.0);
+    thelimiter(index,i) = 0.0;
+    
+    if (rem(i,floor(updateno/20))==0)
+        fprintf(1,'.');
+    end
+    if mod(i,50)==0
+        waitbar_j(i/updateno,hh); 
+    end
+    
+    
+    % reconstruction; first make spectrum odd
+    temp=[femicrophone(:,i);flipud(conj(femicrophone(2:hsupport,i)))];
+    emicrophone(sb:se) = emicrophone(sb:se) + factor * win .* real(ifft(temp));
+
+end
+fprintf(1,'\n');
+
+close(hh);

src/modules/audio_processing/aecm/main/matlab/getBspectrum.m

@@ -0,0 +1,22 @@
+function bspectrum=getBspectrum(ps,threshold,bandfirst,bandlast)
+% function bspectrum=getBspectrum(ps,threshold,bandfirst,bandlast)
+% compute binary spectrum using threshold spectrum as pivot
+% bspectrum = binary spectrum (binary)
+% ps=current power spectrum (float)
+% threshold=threshold spectrum (float)
+% bandfirst = first band considered
+% bandlast = last band considered
+  
+% initialization stuff
+  if( length(ps)<bandlast | bandlast>32 | length(ps)~=length(threshold)) 
+  error('BinDelayEst:spectrum:invalid','Dimensionality error');
+end
+
+% get current binary spectrum
+diff = ps - threshold;
+bspectrum=uint32(0);
+for(i=bandfirst:bandlast)
+  if( diff(i)>0 ) 
+    bspectrum = bitset(bspectrum,i);
+  end
+end

src/modules/audio_processing/aecm/main/matlab/hisser2.m

@@ -0,0 +1,21 @@
+function  bcount=hisser2(bs,bsr,bandfirst,bandlast)
+% function  bcount=hisser(bspectrum,bandfirst,bandlast)
+% histogram for the binary spectra
+% bcount= array of bit counts 
+% bs=binary spectrum (one int32 number each)  
+% bsr=reference binary spectra (one int32 number each)
+% blockSize = histogram over blocksize blocks
+% bandfirst = first band considered
+% bandlast = last band considered
+
+% weight all delays equally
+maxDelay = length(bsr);
+
+% compute counts (two methods; the first works better and is operational)
+bcount=zeros(maxDelay,1);
+for(i=1:maxDelay)
+ % the delay should have low count for low-near&high-far and high-near&low-far
+ bcount(i)= sum(bitget(bitxor(bs,bsr(i)),bandfirst:bandlast));  
+ % the delay should have low count for low-near&high-far (works less well)
+% bcount(i)= sum(bitget(bitand(bsr(i),bitxor(bs,bsr(i))),bandfirst:bandlast));
+end

src/modules/audio_processing/aecm/main/matlab/main2.m

@@ -0,0 +1,19 @@
+
+fid=fopen('aecfar.pcm'); far=fread(fid,'short'); fclose(fid);
+fid=fopen('aecnear.pcm'); mic=fread(fid,'short'); fclose(fid);
+
+%fid=fopen('QA1far.pcm'); far=fread(fid,'short'); fclose(fid);
+%fid=fopen('QA1near.pcm'); mic=fread(fid,'short'); fclose(fid);
+
+start=0 * 8000+1;
+stop= 30 * 8000;
+microphone=mic(start:stop);
+TheFarEnd=far(start:stop);
+avtime=1;
+
+% 16000 to make it compatible with the C-version
+[emicrophone,tdel]=compsup(microphone,TheFarEnd,avtime,16000); 
+
+spclab(8000,TheFarEnd,microphone,emicrophone);
+
+

src/modules/audio_processing/aecm/main/matlab/matlab/AECMobile.m

@@ -0,0 +1,269 @@
+function [femicrophone, aecmStructNew, enerNear, enerFar] = AECMobile(fmicrophone, afTheFarEnd, setupStruct, aecmStruct)
+global NEARENDFFT;
+global F;
+
+aecmStructNew = aecmStruct;
+
+% Magnitude spectrum of near end signal
+afmicrophone = abs(fmicrophone);
+%afmicrophone = NEARENDFFT(setupStruct.currentBlock,:)'/2^F(setupStruct.currentBlock,end);
+% Near end energy level
+ener_orig = afmicrophone'*afmicrophone;
+if( ener_orig == 0)
+    lowlevel = 0.01;
+    afmicrophone = lowlevel*ones(size(afmicrophone));
+end
+%adiff = max(abs(afmicrophone - afTheFarEnd));
+%if (adiff > 0)
+%    disp([setupStruct.currentBlock adiff])
+%end
+
+% Store the near end energy
+%aecmStructNew.enerNear(setupStruct.currentBlock) = log(afmicrophone'*afmicrophone);
+aecmStructNew.enerNear(setupStruct.currentBlock) = log(sum(afmicrophone));
+% Store the far end energy
+%aecmStructNew.enerFar(setupStruct.currentBlock) = log(afTheFarEnd'*afTheFarEnd);
+aecmStructNew.enerFar(setupStruct.currentBlock) = log(sum(afTheFarEnd));
+
+% Update subbands (We currently use all frequency bins, hence .useSubBand is turned off)
+if aecmStructNew.useSubBand
+    internalIndex = 1;
+    for kk=1:setupStruct.subBandLength+1
+        ySubBand(kk) = mean(afmicrophone(internalIndex:internalIndex+setupStruct.numInBand(kk)-1).^aecmStructNew.bandFactor);
+        xSubBand(kk) = mean(afTheFarEnd(internalIndex:internalIndex+setupStruct.numInBand(kk)-1).^aecmStructNew.bandFactor);
+        internalIndex = internalIndex + setupStruct.numInBand(kk);
+    end
+else
+    ySubBand = afmicrophone.^aecmStructNew.bandFactor;
+    xSubBand = afTheFarEnd.^aecmStructNew.bandFactor;
+end
+
+% Estimated echo energy
+if (aecmStructNew.bandFactor == 1)
+    %aecmStructNew.enerEcho(setupStruct.currentBlock) = log((aecmStructNew.H.*xSubBand)'*(aecmStructNew.H.*xSubBand));
+    %aecmStructNew.enerEchoStored(setupStruct.currentBlock) = log((aecmStructNew.HStored.*xSubBand)'*(aecmStructNew.HStored.*xSubBand));
+    aecmStructNew.enerEcho(setupStruct.currentBlock) = log(sum(aecmStructNew.H.*xSubBand));
+    aecmStructNew.enerEchoStored(setupStruct.currentBlock) = log(sum(aecmStructNew.HStored.*xSubBand));
+elseif (aecmStructNew.bandFactor == 2)
+    aecmStructNew.enerEcho(setupStruct.currentBlock) = log(aecmStructNew.H'*xSubBand);
+    aecmStructNew.enerEchoStored(setupStruct.currentBlock) = log(aecmStructNew.HStored'*xSubBand);
+end
+
+% Last 100 blocks of data, used for plotting
+n100 = max(1,setupStruct.currentBlock-99):setupStruct.currentBlock;
+enerError = aecmStructNew.enerNear(n100)-aecmStructNew.enerEcho(n100);
+enerErrorStored = aecmStructNew.enerNear(n100)-aecmStructNew.enerEchoStored(n100);
+
+% Store the far end sub band. This is needed if we use LSE instead of NLMS
+aecmStructNew.X = [xSubBand aecmStructNew.X(:,1:end-1)];
+
+% Update energy levels, which control the VAD
+if ((aecmStructNew.enerFar(setupStruct.currentBlock) < aecmStructNew.energyMin) & (aecmStructNew.enerFar(setupStruct.currentBlock) >= aecmStruct.FAR_ENERGY_MIN))
+    aecmStructNew.energyMin = aecmStructNew.enerFar(setupStruct.currentBlock);
+    %aecmStructNew.energyMin = max(aecmStructNew.energyMin,12);
+    aecmStructNew.energyMin = max(aecmStructNew.energyMin,aecmStruct.FAR_ENERGY_MIN);
+    aecmStructNew.energyLevel = (aecmStructNew.energyMax-aecmStructNew.energyMin)*aecmStructNew.energyThres+aecmStructNew.energyMin;
+    aecmStructNew.energyLevelMSE = (aecmStructNew.energyMax-aecmStructNew.energyMin)*aecmStructNew.energyThresMSE+aecmStructNew.energyMin;
+end
+if (aecmStructNew.enerFar(setupStruct.currentBlock) > aecmStructNew.energyMax)
+    aecmStructNew.energyMax = aecmStructNew.enerFar(setupStruct.currentBlock);
+    aecmStructNew.energyLevel = (aecmStructNew.energyMax-aecmStructNew.energyMin)*aecmStructNew.energyThres+aecmStructNew.energyMin;
+    aecmStructNew.energyLevelMSE = (aecmStructNew.energyMax-aecmStructNew.energyMin)*aecmStructNew.energyThresMSE+aecmStructNew.energyMin;
+end
+
+% Calculate current energy error in near end (estimated echo vs. near end)
+dE = aecmStructNew.enerNear(setupStruct.currentBlock)-aecmStructNew.enerEcho(setupStruct.currentBlock);
+
+%%%%%%%%
+% Calculate step size used in LMS algorithm, based on current far end energy and near end energy error (dE)
+%%%%%%%%
+if setupStruct.stepSize_flag
+    [mu, aecmStructNew] = calcStepSize(aecmStructNew.enerFar(setupStruct.currentBlock), dE, aecmStructNew, setupStruct.currentBlock, 1);
+else
+    mu = 0.25;
+end
+aecmStructNew.muLog(setupStruct.currentBlock) = mu; % Store the step size
+
+% Estimate Echo Spectral Shape
+[U, aecmStructNew.H] = fallerEstimator(ySubBand,aecmStructNew.X,aecmStructNew.H,mu);
+
+%%%%%
+% Determine if we should store or restore the channel
+%%%%%
+if ((setupStruct.currentBlock <= aecmStructNew.convLength) | (~setupStruct.channelUpdate_flag))
+    aecmStructNew.HStored = aecmStructNew.H; % Store what you have after startup
+elseif ((setupStruct.currentBlock > aecmStructNew.convLength) & (setupStruct.channelUpdate_flag))
+    if ((aecmStructNew.enerFar(setupStruct.currentBlock) < aecmStructNew.energyLevelMSE) & (aecmStructNew.enerFar(setupStruct.currentBlock-1) >= aecmStructNew.energyLevelMSE))
+        xxx = aecmStructNew.countMseH;
+        if (xxx > 20)
+            mseStored = mean(abs(aecmStructNew.enerEchoStored(setupStruct.currentBlock-xxx:setupStruct.currentBlock-1)-aecmStructNew.enerNear(setupStruct.currentBlock-xxx:setupStruct.currentBlock-1)));
+            mseLatest = mean(abs(aecmStructNew.enerEcho(setupStruct.currentBlock-xxx:setupStruct.currentBlock-1)-aecmStructNew.enerNear(setupStruct.currentBlock-xxx:setupStruct.currentBlock-1)));
+            %fprintf('Stored: %4f Latest: %4f\n', mseStored, mseLatest) % Uncomment if you want to display the MSE values
+            if ((mseStored < 0.8*mseLatest) & (aecmStructNew.mseHStoredOld < 0.8*aecmStructNew.mseHLatestOld))
+                aecmStructNew.H = aecmStructNew.HStored;
+                fprintf('Restored H at block %d\n',setupStruct.currentBlock)
+            elseif (((0.8*mseStored > mseLatest) & (mseLatest < aecmStructNew.mseHThreshold) & (aecmStructNew.mseHLatestOld < aecmStructNew.mseHThreshold)) | (mseStored == Inf))
+                aecmStructNew.HStored = aecmStructNew.H;
+                fprintf('Stored new H at block %d\n',setupStruct.currentBlock)
+            end
+            aecmStructNew.mseHStoredOld = mseStored;
+            aecmStructNew.mseHLatestOld = mseLatest;
+        end
+    elseif ((aecmStructNew.enerFar(setupStruct.currentBlock) >= aecmStructNew.energyLevelMSE) & (aecmStructNew.enerFar(setupStruct.currentBlock-1) < aecmStructNew.energyLevelMSE))
+        aecmStructNew.countMseH = 1;
+    elseif (aecmStructNew.enerFar(setupStruct.currentBlock) >= aecmStructNew.energyLevelMSE)
+        aecmStructNew.countMseH = aecmStructNew.countMseH + 1;
+    end
+end
+
+%%%%%
+% Check delay (calculate the delay offset (if we can))
+% The algorithm is not tuned and should be used with care. It runs separately from Bastiaan's algorithm.
+%%%%%
+yyy = 31; % Correlation buffer length (currently unfortunately hard coded)
+dxxx = 25; % Maximum offset (currently unfortunately hard coded)
+if (setupStruct.currentBlock > aecmStructNew.convLength)
+    if (aecmStructNew.enerFar(setupStruct.currentBlock-(yyy+2*dxxx-1):setupStruct.currentBlock) > aecmStructNew.energyLevelMSE)
+        for xxx = -dxxx:dxxx
+            aecmStructNew.delayLatestS(xxx+dxxx+1) = sum(sign(aecmStructNew.enerEcho(setupStruct.currentBlock-(yyy+dxxx-xxx)+1:setupStruct.currentBlock+xxx-dxxx)-mean(aecmStructNew.enerEcho(setupStruct.currentBlock-(yyy++dxxx-xxx)+1:setupStruct.currentBlock+xxx-dxxx))).*sign(aecmStructNew.enerNear(setupStruct.currentBlock-yyy-dxxx+1:setupStruct.currentBlock-dxxx)-mean(aecmStructNew.enerNear(setupStruct.currentBlock-yyy-dxxx+1:setupStruct.currentBlock-dxxx))));
+        end
+        aecmStructNew.newDelayCurve = 1;
+    end
+end
+if ((setupStruct.currentBlock > 2*aecmStructNew.convLength) & ~rem(setupStruct.currentBlock,yyy*2) & aecmStructNew.newDelayCurve)
+    [maxV,maxP] = max(aecmStructNew.delayLatestS);
+    if ((maxP > 2) & (maxP < 2*dxxx))
+        maxVLeft = aecmStructNew.delayLatestS(max(1,maxP-4));
+        maxVRight = aecmStructNew.delayLatestS(min(2*dxxx+1,maxP+4));
+        %fprintf('Max %d, Left %d, Right %d\n',maxV,maxVLeft,maxVRight) % Uncomment if you want to see max value
+        if ((maxV > 24) & (maxVLeft < maxV - 10)  & (maxVRight < maxV - 10))
+            aecmStructNew.feedbackDelay = maxP-dxxx-1;
+            aecmStructNew.newDelayCurve = 0;
+            aecmStructNew.feedbackDelayUpdate = 1;
+            fprintf('Feedback Update at block %d\n',setupStruct.currentBlock)
+        end
+    end
+end
+% End of "Check delay"
+%%%%%%%%
+
+%%%%%
+% Calculate suppression gain, based on far end energy and near end energy error (dE)
+if (setupStruct.supGain_flag)
+    [gamma_echo, aecmStructNew.cntIn, aecmStructNew.cntOut] = calcFilterGain(aecmStructNew.enerFar(setupStruct.currentBlock), dE, aecmStructNew, setupStruct.currentBlock, aecmStructNew.convLength, aecmStructNew.cntIn, aecmStructNew.cntOut);
+else
+    gamma_echo = 1;
+end
+aecmStructNew.gammaLog(setupStruct.currentBlock) = gamma_echo; % Store the gain
+gamma_use = gamma_echo;
+
+% Use the stored channel
+U = aecmStructNew.HStored.*xSubBand;
+
+% compute Wiener filter and suppressor function
+Iy = find(ySubBand);
+subBandFilter = zeros(size(ySubBand));
+if (aecmStructNew.bandFactor == 2)
+    subBandFilter(Iy) = (1 - gamma_use*sqrt(U(Iy)./ySubBand(Iy))); % For Faller
+else
+    subBandFilter(Iy) = (1 - gamma_use*(U(Iy)./ySubBand(Iy))); % For COV
+end
+ix0 = find(subBandFilter < 0);   % bounding trick 1
+subBandFilter(ix0) = 0;
+ix0 = find(subBandFilter > 1);   % bounding trick 1
+subBandFilter(ix0) = 1;
+
+% Interpolate back to normal frequency bins if we use sub bands
+if aecmStructNew.useSubBand
+    thefilter = interp1(setupStruct.centerFreq,subBandFilter,linspace(0,setupStruct.samplingfreq/2,setupStruct.hsupport1)','nearest');
+    testfilter = interp1(setupStruct.centerFreq,subBandFilter,linspace(0,setupStruct.samplingfreq/2,1000),'nearest');
+    thefilter(end) = subBandFilter(end);
+    
+    internalIndex = 1;
+    for kk=1:setupStruct.subBandLength+1
+        internalIndex:internalIndex+setupStruct.numInBand(kk)-1;
+        thefilter(internalIndex:internalIndex+setupStruct.numInBand(kk)-1) = subBandFilter(kk);
+        internalIndex = internalIndex + setupStruct.numInBand(kk);
+    end
+else
+    thefilter = subBandFilter;
+    testfilter = subBandFilter;
+end
+
+% Bound the filter
+ix0 = find(thefilter < setupStruct.de_echo_bound);   % bounding trick 1
+thefilter(ix0) = setupStruct.de_echo_bound;     % bounding trick 2
+ix0 = find(thefilter > 1);   % bounding in reasonable range
+thefilter(ix0) = 1;
+
+%%%%
+% NLP
+%%%%
+thefmean = mean(thefilter(8:16));
+if (thefmean < 1)
+    disp('');
+end
+aecmStructNew.runningfmean = setupStruct.nl_alpha*aecmStructNew.runningfmean + (1-setupStruct.nl_alpha)*thefmean;
+slope0 = 1.0/(1.0 - setupStruct.nlSeverity); %
+thegain = max(0.0, min(1.0, slope0*(aecmStructNew.runningfmean - setupStruct.nlSeverity)));
+if ~setupStruct.nlp_flag
+    thegain = 1;
+end
+% END NONLINEARITY
+thefilter = thegain*thefilter;
+
+%%%%
+% The suppression
+%%%%
+femicrophone = fmicrophone .* thefilter;
+% Store the output energy (used for plotting)
+%aecmStructNew.enerOut(setupStruct.currentBlock) = log(abs(femicrophone)'*abs(femicrophone));
+aecmStructNew.enerOut(setupStruct.currentBlock) = log(sum(abs(femicrophone)));
+
+if aecmStructNew.plotIt
+    figure(13)
+    subplot(311)
+    %plot(n100,enerFar(n100),'b-',n100,enerNear(n100),'k--',n100,enerEcho(n100),'r-',[n100(1) n100(end)],[1 1]*vadThNew,'b:',[n100(1) n100(end)],[1 1]*((energyMax-energyMin)/4+energyMin),'r-.',[n100(1) n100(end)],[1 1]*vadNearThNew,'g:',[n100(1) n100(end)],[1 1]*energyMax,'r-.',[n100(1) n100(end)],[1 1]*energyMin,'r-.','LineWidth',2)
+    plot(n100,aecmStructNew.enerFar(n100),'b-',n100,aecmStructNew.enerNear(n100),'k--',n100,aecmStructNew.enerOut(n100),'r-.',n100,aecmStructNew.enerEcho(n100),'r-',n100,aecmStructNew.enerEchoStored(n100),'c-',[n100(1) n100(end)],[1 1]*((aecmStructNew.energyMax-aecmStructNew.energyMin)/4+aecmStructNew.energyMin),'g-.',[n100(1) n100(end)],[1 1]*aecmStructNew.energyMax,'g-.',[n100(1) n100(end)],[1 1]*aecmStructNew.energyMin,'g-.','LineWidth',2)
+    %title(['Frame ',int2str(i),' av ',int2str(setupStruct.updateno),' State = ',int2str(speechState),' \mu = ',num2str(mu)])
+    title(['\gamma = ',num2str(gamma_echo),' \mu = ',num2str(mu)])
+    subplot(312)
+    %plot(n100,enerError,'b-',[n100(1) n100(end)],[1 1]*vadNearTh,'r:',[n100(1) n100(end)],[-1.5 -1.5]*vadNearTh,'r:','LineWidth',2)
+    %plot(n100,enerError,'b-',[n100(1) n100(end)],[1 1],'r:',[n100(1) n100(end)],[-2 -2],'r:','LineWidth',2)
+    plot(n100,enerError,'b-',n100,enerErrorStored,'c-',[n100(1) n100(end)],[1 1]*aecmStructNew.varMean,'k--',[n100(1) n100(end)],[1 1],'r:',[n100(1) n100(end)],[-2 -2],'r:','LineWidth',2)
+    % Plot mu
+    %plot(n100,log2(aecmStructNew.muLog(n100)),'b-','LineWidth',2)
+    %plot(n100,log2(aecmStructNew.HGain(n100)),'b-',[n100(1) n100(end)],[1 1]*log2(sum(aecmStructNew.HStored)),'r:','LineWidth',2)
+    title(['Block ',int2str(setupStruct.currentBlock),' av ',int2str(setupStruct.updateno)])
+    subplot(313)
+    %plot(n100,enerVar(n100),'b-',[n100(1) n100(end)],[1 1],'r:',[n100(1) n100(end)],[-2 -2],'r:','LineWidth',2)
+    %plot(n100,enerVar(n100),'b-','LineWidth',2)
+    % Plot correlation curve
+
+    %plot(-25:25,aecmStructNew.delayStored/max(aecmStructNew.delayStored),'c-',-25:25,aecmStructNew.delayLatest/max(aecmStructNew.delayLatest),'r-',-25:25,(max(aecmStructNew.delayStoredS)-aecmStructNew.delayStoredS)/(max(aecmStructNew.delayStoredS)-min(aecmStructNew.delayStoredS)),'c:',-25:25,(max(aecmStructNew.delayLatestS)-aecmStructNew.delayLatestS)/(max(aecmStructNew.delayLatestS)-min(aecmStructNew.delayLatestS)),'r:','LineWidth',2)
+    %plot(-25:25,aecmStructNew.delayStored,'c-',-25:25,aecmStructNew.delayLatest,'r-',-25:25,(max(aecmStructNew.delayStoredS)-aecmStructNew.delayStoredS)/(max(aecmStructNew.delayStoredS)-min(aecmStructNew.delayStoredS)),'c:',-25:25,(max(aecmStructNew.delayLatestS)-aecmStructNew.delayLatestS)/(max(aecmStructNew.delayLatestS)-min(aecmStructNew.delayLatestS)),'r:','LineWidth',2)
+    %plot(-25:25,aecmStructNew.delayLatest,'r-',-25:25,(50-aecmStructNew.delayLatestS)/100,'r:','LineWidth',2)
+    plot(-25:25,aecmStructNew.delayLatestS,'r:','LineWidth',2)
+    %plot(-25:25,aecmStructNew.delayStored,'c-',-25:25,aecmStructNew.delayLatest,'r-','LineWidth',2)
+    plot(0:32,aecmStruct.HStored,'bo-','LineWidth',2)
+    %title(['\gamma | In = ',int2str(aecmStructNew.muStruct.countInInterval),' | Out High = ',int2str(aecmStructNew.muStruct.countOutHighInterval),' | Out Low = ',int2str(aecmStructNew.muStruct.countOutLowInterval)])
+    pause(1)
+    %if ((setupStruct.currentBlock == 860) | (setupStruct.currentBlock == 420) | (setupStruct.currentBlock == 960))
+    if 0%(setupStruct.currentBlock == 960)
+        figure(60)
+        plot(n100,aecmStructNew.enerNear(n100),'k--',n100,aecmStructNew.enerEcho(n100),'k:','LineWidth',2)
+        legend('Near End','Estimated Echo')
+        title('Signal Energy witH offset compensation')
+        figure(61)
+        subplot(211)
+        stem(sign(aecmStructNew.enerNear(n100)-mean(aecmStructNew.enerNear(n100))))
+        title('Near End Energy Pattern (around mean value)')
+        subplot(212)
+        stem(sign(aecmStructNew.enerEcho(n100)-mean(aecmStructNew.enerEcho(n100))))
+        title('Estimated Echo Energy Pattern (around mean value)')
+        pause
+    end
+    drawnow%,pause
+elseif ~rem(setupStruct.currentBlock,100)
+    fprintf('Block %d of %d\n',setupStruct.currentBlock,setupStruct.updateno)
+end

src/modules/audio_processing/aecm/main/matlab/matlab/align.m

@@ -0,0 +1,98 @@
+function [delayStructNew] = align(xf, yf, delayStruct, i, trueDelay);
+
+%%%%%%%
+% Bastiaan's algorithm copied
+%%%%%%%
+Ap500 = [1.00, -4.95, 9.801, -9.70299, 4.80298005, -0.9509900499];
+Bp500 = [0.662743088639636, -2.5841655608125, 3.77668102146288, -2.45182477425154, 0.596566274575251, 0.0];
+Ap200 = [1.00, -4.875, 9.50625, -9.26859375, 4.518439453125, -0.881095693359375];
+Bp200 = [0.862545460994275, -3.2832804496114, 4.67892032308828, -2.95798023879133, 0.699796870041299, 0.0];
+
+oldMethod = 1; % Turn on or off the old method. The new one is Bastiaan's August 2008 updates
+THReSHoLD = 2.0; % ADJUSTABLE threshold factor; 4.0 seems good
+%%%%%%%%%%%%%%%%%%%
+% use log domain (showed improved performance)
+xxf = sqrt(real(xf.*conj(xf))+1e-20);
+yyf = sqrt(real(yf.*conj(yf))+1e-20);
+delayStruct.sxAll2(:,i) = 20*log10(xxf);
+delayStruct.syAll2(:,i) = 20*log10(yyf);
+
+mD = min(i-1,delayStruct.maxDelayb);
+if oldMethod
+    factor = 1.0;
+    histLenb = 250;
+    xthreshold = factor*median(delayStruct.sxAll2(:,i-mD:i),2);
+    ythreshold = factor*median(delayStruct.syAll2(:,i-mD:i),2);
+else
+    xthreshold = sum(delayStruct.sxAll2(:,i-mD:i),2)/(delayStruct.maxDelayb+1);
+    
+    [yout, delayStruct.z200] = filter(Bp200, Ap200, delayStruct.syAll2(:,i), delayStruct.z200, 2);
+    yout = yout/(delayStruct.maxDelayb+1);
+    ythreshold = mean(delayStruct.syAll2(:,i-mD:i),2);
+    ythreshold = yout;
+end
+
+delayStruct.bxspectrum(i) = getBspectrum(delayStruct.sxAll2(:,i), xthreshold, delayStruct.bandfirst, delayStruct.bandlast);
+delayStruct.byspectrum(i) = getBspectrum(delayStruct.syAll2(:,i), ythreshold, delayStruct.bandfirst, delayStruct.bandlast);
+
+delayStruct.bxhist(end-mD:end) = delayStruct.bxspectrum(i-mD:i);
+
+delayStruct.bcount(:,i) = hisser2(delayStruct.byspectrum(i), flipud(delayStruct.bxhist), delayStruct.bandfirst, delayStruct.bandlast);
+[delayStruct.fout(:,i), delayStruct.z500] = filter(Bp500, Ap500, delayStruct.bcount(:,i), delayStruct.z500, 2);
+if oldMethod
+    %delayStruct.new(:,i) = sum(delayStruct.bcount(:,max(1,i-histLenb+1):i),2); % using the history range
+    tmpVec = [delayStruct.fout(1,i)*ones(2,1); delayStruct.fout(:,i); delayStruct.fout(end,i)*ones(2,1)]; % using the history range
+    tmpVec = filter(ones(1,5), 1, tmpVec);
+    delayStruct.new(:,i) = tmpVec(5:end);
+    %delayStruct.new(:,i) = delayStruct.fout(:,i); % using the history range
+else
+    [delayStruct.fout(:,i), delayStruct.z500] = filter(Bp500, Ap500, delayStruct.bcount(:,i), delayStruct.z500, 2);
+    % NEW CODE
+    delayStruct.new(:,i) = filter([-1,-2,1,4,1,-2,-1], 1, delayStruct.fout(:,i)); %remv smth component
+    delayStruct.new(1:end-3,i) = delayStruct.new(1+3:end,i);
+    delayStruct.new(1:6,i) = 0.0;
+    delayStruct.new(end-6:end,i) = 0.0;  % ends are no good
+end
+[valuen, tempdelay] = min(delayStruct.new(:,i));  % find minimum
+if oldMethod
+    threshold = valuen + (max(delayStruct.new(:,i)) - valuen)/4;
+    thIndex = find(delayStruct.new(:,i) <= threshold);
+    if (i > 1)
+        delayDiff = abs(delayStruct.delay(i-1)-tempdelay+1);
+        if (delayStruct.oneGoodEstimate & (max(diff(thIndex)) > 1) & (delayDiff < 10))
+            % We consider this minimum to be significant, hence update the delay
+            delayStruct.delay(i) = tempdelay;
+        elseif (~delayStruct.oneGoodEstimate & (max(diff(thIndex)) > 1))
+            delayStruct.delay(i) = tempdelay;
+            if (i > histLenb)
+                delayStruct.oneGoodEstimate = 1;
+            end
+        else
+            delayStruct.delay(i) = delayStruct.delay(i-1);
+        end
+    else
+        delayStruct.delay(i) = tempdelay;
+    end
+else
+    threshold = THReSHoLD*std(delayStruct.new(:,i));   % set updata threshold 
+    if ((-valuen > threshold) | (i < delayStruct.smlength)) % see if you want to update delay
+        delayStruct.delay(i) = tempdelay;
+    else
+        delayStruct.delay(i) = delayStruct.delay(i-1);
+    end
+    % END NEW CODE
+end
+delayStructNew = delayStruct;
+
+% administrative and plotting stuff
+if( 0)
+    figure(10);
+    plot([1:length(delayStructNew.new(:,i))],delayStructNew.new(:,i),trueDelay*[1 1],[min(delayStructNew.new(:,i)),max(delayStructNew.new(:,i))],'r',[1 length(delayStructNew.new(:,i))],threshold*[1 1],'r:', 'LineWidth',2);
+    %plot([1:length(delayStructNew.bcount(:,i))],delayStructNew.bcount(:,i),trueDelay*[1 1],[min(delayStructNew.bcount(:,i)),max(delayStructNew.bcount(:,i))],'r','LineWidth',2);
+    %plot([thedelay,thedelay],[min(fcount(:,i)),max(fcount(:,i))],'r');
+    %title(sprintf('bin count and known delay at time %5.1f s\n',(i-1)*(support/(fs*oversampling))));
+    title(delayStructNew.oneGoodEstimate)
+    xlabel('delay in frames');
+    %hold off;
+    drawnow
+end

src/modules/audio_processing/aecm/main/matlab/matlab/calcFilterGain.m

@@ -0,0 +1,88 @@
+function [gam, cntIn2, cntOut2] = calcFilterGain(energy, dE, aecmStruct, t, T, cntIn, cntOut)
+
+defaultLevel = 1.2;
+cntIn2 = cntIn;
+cntOut2 = cntOut;
+if (t < T)
+    gam = 1;
+else
+    dE1 = -5;
+    dE2 = 1;
+    gamMid = 0.2;
+    gam = max(0,min((energy - aecmStruct.energyMin)/(aecmStruct.energyLevel - aecmStruct.energyMin), 1-(1-gamMid)*(aecmStruct.energyMax-energy)/(aecmStruct.energyMax-aecmStruct.energyLevel)));
+    
+    dEOffset = -0.5;
+    dEWidth = 1.5;
+    %gam2 = max(1,2-((dE-dEOffset)/(dE2-dEOffset)).^2);
+    gam2 = 1+(abs(dE-dEOffset)<(dE2-dEOffset));
+    
+    gam = gam*gam2;
+    
+    
+    if (energy < aecmStruct.energyLevel)
+        gam = 0;
+    else
+        gam = defaultLevel;
+    end
+    dEVec = aecmStruct.enerNear(t-63:t)-aecmStruct.enerEcho(t-63:t);
+    %dEVec = aecmStruct.enerNear(t-20:t)-aecmStruct.enerEcho(t-20:t);
+    numCross = 0;
+    currentState = 0;
+    for ii=1:64
+        if (currentState == 0)
+            currentState = (dEVec(ii) > dE2) - (dEVec(ii) < -2);
+        elseif ((currentState == 1) & (dEVec(ii) < -2))
+            numCross = numCross + 1;
+            currentState = -1;
+        elseif ((currentState == -1) & (dEVec(ii) > dE2))
+            numCross = numCross + 1;
+            currentState = 1;
+        end
+    end
+    gam = max(0, gam - numCross/25);
+    gam = 1;
+    
+    ener_A = 1;
+    ener_B = 0.8;
+    ener_C = aecmStruct.energyLevel + (aecmStruct.energyMax-aecmStruct.energyLevel)/5;
+    dE_A = 4;%2;
+    dE_B = 3.6;%1.8;
+    dE_C = 0.9*dEWidth;
+    dE_D = 1;
+    timeFactorLength = 10;
+    ddE = abs(dE-dEOffset);
+    if (energy < aecmStruct.energyLevel)
+        gam = 0;
+    else
+        gam = 1;
+        gam2 = max(0, min(ener_B*(energy-aecmStruct.energyLevel)/(ener_C-aecmStruct.energyLevel), ener_B+(ener_A-ener_B)*(energy-ener_C)/(aecmStruct.energyMax-ener_C)));
+        if (ddE < dEWidth)
+            % Update counters
+            cntIn2 = cntIn2 + 1;
+            if (cntIn2 > 2)
+                cntOut2 = 0;
+            end
+            gam3 = max(dE_D, min(dE_A-(dE_A-dE_B)*(ddE/dE_C), dE_D+(dE_B-dE_D)*(dEWidth-ddE)/(dEWidth-dE_C)));
+            gam3 = dE_A;
+        else
+            % Update counters
+            cntOut2 = cntOut2 + 1;
+            if (cntOut2 > 2)
+                cntIn2 = 0;
+            end
+            %gam2 = 1;
+            gam3 = dE_D;
+        end
+        timeFactor = min(1, cntIn2/timeFactorLength);
+        gam = gam*(1-timeFactor) + timeFactor*gam2*gam3;
+    end
+    %gam = gam/floor(numCross/2+1);
+end
+if isempty(gam)
+    numCross
+    timeFactor
+    cntIn2
+    cntOut2
+    gam2
+    gam3
+end

src/modules/audio_processing/aecm/main/matlab/matlab/calcStepSize.m

@@ -0,0 +1,105 @@
+function [mu, aecmStructNew] = calcStepSize(energy, dE, aecmStruct, t, logscale)
+
+if (nargin < 4)
+    t = 1;
+    logscale = 1;
+elseif (nargin == 4)
+    logscale = 1;
+end
+T = aecmStruct.convLength;
+
+if logscale
+    currentMuMax = aecmStruct.MU_MIN + (aecmStruct.MU_MAX-aecmStruct.MU_MIN)*min(t,T)/T;
+    if (aecmStruct.energyMin >= aecmStruct.energyMax)
+        mu = aecmStruct.MU_MIN;
+    else
+        mu = (energy - aecmStruct.energyMin)/(aecmStruct.energyMax - aecmStruct.energyMin)*(currentMuMax-aecmStruct.MU_MIN) + aecmStruct.MU_MIN;
+    end
+    mu = 2^mu;
+    if (energy < aecmStruct.energyLevel)
+        mu = 0;
+    end
+else
+    muMin = 0;
+    muMax = 0.5;
+    currentMuMax = muMin + (muMax-muMin)*min(t,T)/T;
+    if (aecmStruct.energyMin >= aecmStruct.energyMax)
+        mu = muMin;
+    else
+        mu = (energy - aecmStruct.energyMin)/(aecmStruct.energyMax - aecmStruct.energyMin)*(currentMuMax-muMin) + muMin;
+    end
+end
+dE2 = 1;
+dEOffset = -0.5;
+offBoost = 5;
+if (mu > 0)
+    if (abs(dE-aecmStruct.ENERGY_DEV_OFFSET) > aecmStruct.ENERGY_DEV_TOL)
+        aecmStruct.muStruct.countInInterval = 0;
+    else
+        aecmStruct.muStruct.countInInterval = aecmStruct.muStruct.countInInterval + 1;
+    end
+    if (dE < aecmStruct.ENERGY_DEV_OFFSET - aecmStruct.ENERGY_DEV_TOL)
+        aecmStruct.muStruct.countOutLowInterval = aecmStruct.muStruct.countOutLowInterval + 1;
+    else
+        aecmStruct.muStruct.countOutLowInterval = 0;
+    end
+    if (dE > aecmStruct.ENERGY_DEV_OFFSET + aecmStruct.ENERGY_DEV_TOL)
+        aecmStruct.muStruct.countOutHighInterval = aecmStruct.muStruct.countOutHighInterval + 1;
+    else
+        aecmStruct.muStruct.countOutHighInterval = 0;
+    end
+end
+muVar = 2^min(-3,5/50*aecmStruct.muStruct.countInInterval-3);
+muOff = 2^max(offBoost,min(0,offBoost*(aecmStruct.muStruct.countOutLowInterval-aecmStruct.muStruct.minOutLowInterval)/(aecmStruct.muStruct.maxOutLowInterval-aecmStruct.muStruct.minOutLowInterval)));
+
+muLow = 1/64;
+muVar = 1;
+if (t < 2*T)
+    muDT = 1;
+    muVar = 1;
+    mdEVec = 0;
+    numCross = 0;
+else
+    muDT = min(1,max(muLow,1-(1-muLow)*(dE-aecmStruct.ENERGY_DEV_OFFSET)/aecmStruct.ENERGY_DEV_TOL));
+    dEVec = aecmStruct.enerNear(t-63:t)-aecmStruct.enerEcho(t-63:t);
+    %dEVec = aecmStruct.enerNear(t-20:t)-aecmStruct.enerEcho(t-20:t);
+    numCross = 0;
+    currentState = 0;
+    for ii=1:64
+        if (currentState == 0)
+            currentState = (dEVec(ii) > dE2) - (dEVec(ii) < -2);
+        elseif ((currentState == 1) & (dEVec(ii) < -2))
+            numCross = numCross + 1;
+            currentState = -1;
+        elseif ((currentState == -1) & (dEVec(ii) > dE2))
+            numCross = numCross + 1;
+            currentState = 1;
+        end
+    end
+            
+    %logicDEVec = (dEVec > dE2) - (dEVec < -2);
+    %numCross = sum(abs(diff(logicDEVec)));
+    %mdEVec = mean(abs(dEVec-dEOffset));
+    %mdEVec = mean(abs(dEVec-mean(dEVec)));
+    %mdEVec = max(dEVec)-min(dEVec);
+    %if (mdEVec > 4)%1.5)
+    %    muVar = 0;
+    %end
+    muVar = 2^(-floor(numCross/2));
+    muVar = 2^(-numCross);
+end
+%muVar = 1;
+
+
+% if (eStd > (dE2-dEOffset))
+%     muVar = 1/8;
+% else
+%     muVar = 1;
+% end
+
+%mu = mu*muDT*muVar*muOff;
+mu = mu*muDT*muVar;
+mu = min(mu,0.25);
+aecmStructNew = aecmStruct;
+%aecmStructNew.varMean = mdEVec;
+aecmStructNew.varMean = numCross;

src/modules/audio_processing/aecm/main/matlab/matlab/fallerEstimator.m

@@ -0,0 +1,42 @@
+function [U, Hnew] = fallerEstimator(Y, X, H, mu)
+
+% Near end signal is stacked frame by frame columnwise in matrix Y and far end in X
+%
+% Possible estimation procedures are
+% 1) LSE
+% 2) NLMS
+% 3) Separated numerator and denomerator filters
+regParam = 1;
+[numFreqs, numFrames] = size(Y);
+[numFreqs, Q] = size(X);
+U = zeros(numFreqs, 1);
+
+if ((nargin == 3) | (nargin == 5))
+    dtd = 0;
+end
+if (nargin == 4)
+    dtd = H;
+end
+Emax = 7;
+dEH = Emax-sum(sum(H));
+nu = 2*mu;
+% if (nargin < 5)
+%     H = zeros(numFreqs, Q);
+%     for kk = 1:numFreqs
+%         Xmatrix = hankel(X(kk,1:Q),X(kk,Q:end));
+%         y = Y(kk,1:end-Q+1)';
+%         H(kk,:) = (y'*Xmatrix')*inv(Xmatrix*Xmatrix'+regParam);
+%         U(kk,1) = H(kk,:)*Xmatrix(:,1);
+%     end
+% else
+    for kk = 1:numFreqs
+        x = X(kk,1:Q)';
+        y = Y(kk,1);
+        Htmp = mu*(y-H(kk,:)*x)/(x'*x+regParam)*x;
+        %Htmp = (mu*(y-H(kk,:)*x)/(x'*x+regParam) - nu/dEH)*x;
+        H(kk,:) = H(kk,:) + Htmp';
+        U(kk,1) = H(kk,:)*x;
+    end
+% end
+
+Hnew = H;

src/modules/audio_processing/aecm/main/matlab/matlab/getBspectrum.m

@@ -0,0 +1,22 @@
+function bspectrum=getBspectrum(ps,threshold,bandfirst,bandlast)
+% function bspectrum=getBspectrum(ps,threshold,bandfirst,bandlast)
+% compute binary spectrum using threshold spectrum as pivot
+% bspectrum = binary spectrum (binary)
+% ps=current power spectrum (float)
+% threshold=threshold spectrum (float)
+% bandfirst = first band considered
+% bandlast = last band considered
+  
+% initialization stuff
+  if( length(ps)<bandlast | bandlast>32 | length(ps)~=length(threshold)) 
+  error('BinDelayEst:spectrum:invalid','Dimensionality error');
+end
+
+% get current binary spectrum
+diff = ps - threshold;
+bspectrum=uint32(0);
+for(i=bandfirst:bandlast)
+  if( diff(i)>0 ) 
+    bspectrum = bitset(bspectrum,i);
+  end
+end

src/modules/audio_processing/aecm/main/matlab/matlab/hisser2.m

@@ -0,0 +1,21 @@
+function  bcount=hisser2(bs,bsr,bandfirst,bandlast)
+% function  bcount=hisser(bspectrum,bandfirst,bandlast)
+% histogram for the binary spectra
+% bcount= array of bit counts 
+% bs=binary spectrum (one int32 number each)  
+% bsr=reference binary spectra (one int32 number each)
+% blockSize = histogram over blocksize blocks
+% bandfirst = first band considered
+% bandlast = last band considered
+
+% weight all delays equally
+maxDelay = length(bsr);
+
+% compute counts (two methods; the first works better and is operational)
+bcount=zeros(maxDelay,1);
+for(i=1:maxDelay)
+ % the delay should have low count for low-near&high-far and high-near&low-far
+ bcount(i)= sum(bitget(bitxor(bs,bsr(i)),bandfirst:bandlast));  
+ % the delay should have low count for low-near&high-far (works less well)
+% bcount(i)= sum(bitget(bitand(bsr(i),bitxor(bs,bsr(i))),bandfirst:bandlast));
+end

src/modules/audio_processing/aecm/main/matlab/matlab/mainProgram.m

@@ -0,0 +1,283 @@
+useHTC = 1; % Set this if you want to run a single file and set file names below. Otherwise use simEnvironment to run from several scenarios in a row
+delayCompensation_flag = 0; % Set this flag to one if you want to turn on the delay compensation/enhancement
+global FARENDFFT;
+global NEARENDFFT;
+global F;
+
+if useHTC
+%    fid=fopen('./htcTouchHd/nb/aecFar.pcm'); xFar=fread(fid,'short'); fclose(fid);
+%    fid=fopen('./htcTouchHd/nb/aecNear.pcm'); yNear=fread(fid,'short'); fclose(fid);
+%    fid=fopen('./samsungBlackjack/nb/aecFar.pcm'); xFar=fread(fid,'short'); fclose(fid);
+%    fid=fopen('./samsungBlackjack/nb/aecNear.pcm'); yNear=fread(fid,'short'); fclose(fid);
+%     fid=fopen('aecFarPoor.pcm'); xFar=fread(fid,'short'); fclose(fid);
+%     fid=fopen('aecNearPoor.pcm'); yNear=fread(fid,'short'); fclose(fid);
+%     fid=fopen('out_aes.pcm'); outAES=fread(fid,'short'); fclose(fid);
+   fid=fopen('aecFar4.pcm'); xFar=fread(fid,'short'); fclose(fid);
+   fid=fopen('aecNear4.pcm'); yNear=fread(fid,'short'); fclose(fid);
+    yNearSpeech = zeros(size(xFar));
+     fs = 8000;
+     frameSize = 64;
+%     frameSize = 128;
+     fs = 16000;
+%     frameSize = 256;
+%F = load('fftValues.txt');
+%FARENDFFT = F(:,1:33);
+%NEARENDFFT = F(:,34:66);
+
+else
+    loadFileFar = [speakerType, '_s_',scenario,'_far_b.wav'];
+    [xFar,fs,nbits] = wavread(loadFileFar);
+    xFar = xFar*2^(nbits-1);
+    loadFileNear = [speakerType, '_s_',scenario,'_near_b.wav'];
+    [yNear,fs,nbits] = wavread(loadFileNear);
+    yNear = yNear*2^(nbits-1);
+    loadFileNearSpeech = [speakerType, '_s_',scenario,'_nearSpeech_b.wav'];
+    [yNearSpeech,fs,nbits] = wavread(loadFileNearSpeech);
+    yNearSpeech = yNearSpeech*2^(nbits-1);
+    frameSize = 256;
+end
+
+dtRegions = [];
+
+% General settings for the AECM
+setupStruct = struct(...
+    'stepSize_flag', 1,...      % This flag turns on the step size calculation. If turned off, mu = 0.25.
+    'supGain_flag', 0,...       % This flag turns on the suppression gain calculation. If turned off, gam = 1.
+    'channelUpdate_flag', 0,... % This flag turns on the channel update. If turned off, H is updated for convLength and then kept constant.
+    'nlp_flag', 0,...           % Turn on/off NLP
+    'withVAD_flag', 0,...           % Turn on/off NLP
+    'useSubBand', 0,...         % Set to 1 if to use subBands
+    'useDelayEstimation', 1,... % Set to 1 if to use delay estimation
+    'support', frameSize,...    % # of samples per frame
+    'samplingfreq',fs,...       % Sampling frequency
+    'oversampling', 2,...       % Overlap between blocks/frames
+    'updatel', 0,...            % # of samples between blocks
+    'hsupport1', 0,...          % # of bins in frequency domain
+    'factor', 0,...             % synthesis window amplification
+    'tlength', 0,...            % # of samples of entire file
+    'updateno', 0,...           % # of updates
+    'nb', 1,...                 % # of blocks
+    'currentBlock', 0,...       %
+    'win', zeros(frameSize,1),...% Window to apply for fft and synthesis
+    'avtime', 1,...             % Time (in sec.) to perform averaging
+    'estLen', 0,...             % Averaging in # of blocks
+    'A_GAIN', 10.0,...          % 
+    'suppress_overdrive', 1.0,...   % overdrive factor for suppression 1.4 is good
+    'gamma_echo', 1.0,...       % same as suppress_overdrive but at different place
+    'de_echo_bound', 0.0,...    %
+    'nl_alpha', 0.4,...         % memory; seems not very critical
+    'nlSeverity', 0.2,...         % nonlinearity severity: 0 does nothing; 1 suppresses all
+    'numInBand', [],...         % # of frequency bins in resp. subBand
+    'centerFreq', [],...        % Center frequency of resp. subBand
+    'dtRegions', dtRegions,...  % Regions where we have DT
+    'subBandLength', frameSize/2);%All bins
+    %'subBandLength', 11);       %Something's wrong when subBandLength even
+    %'nl_alpha', 0.8,...         % memory; seems not very critical
+
+delayStruct = struct(...
+    'bandfirst', 8,...
+    'bandlast', 25,...
+    'smlength', 600,...
+    'maxDelay', 0.4,...
+    'oneGoodEstimate', 0,...
+    'delayAdjust', 0,...
+    'maxDelayb', 0);
+% More parameters in delayStruct are constructed in "updateSettings" below
+
+% Make struct settings
+[setupStruct, delayStruct] = updateSettings(yNear, xFar, setupStruct, delayStruct);
+setupStruct.numInBand = ones(setupStruct.hsupport1,1);
+
+Q = 1; % Time diversity in channel
+% General settings for the step size calculation
+muStruct = struct(...
+    'countInInterval', 0,...
+    'countOutHighInterval', 0,...
+    'countOutLowInterval', 0,...
+    'minInInterval', 50,...
+    'minOutHighInterval', 10,...
+    'minOutLowInterval', 10,...
+    'maxOutLowInterval', 50);
+% General settings for the AECM
+aecmStruct = struct(...
+    'plotIt', 0,... % Set to 0 to turn off plotting
+    'useSubBand', 0,...
+    'bandFactor', 1,...
+    'H', zeros(setupStruct.subBandLength+1,Q),...
+    'HStored', zeros(setupStruct.subBandLength+1,Q),...
+    'X', zeros(setupStruct.subBandLength+1,Q),...
+    'energyThres', 0.28,...
+    'energyThresMSE', 0.4,...
+    'energyMin', inf,...
+    'energyMax', -inf,...
+    'energyLevel', 0,...
+    'energyLevelMSE', 0,...
+    'convLength', 100,...
+    'gammaLog', ones(setupStruct.updateno,1),...
+    'muLog', ones(setupStruct.updateno,1),...
+    'enerFar', zeros(setupStruct.updateno,1),...
+    'enerNear', zeros(setupStruct.updateno,1),...
+    'enerEcho', zeros(setupStruct.updateno,1),...
+    'enerEchoStored', zeros(setupStruct.updateno,1),...
+    'enerOut', zeros(setupStruct.updateno,1),...
+    'runningfmean', 0,...
+    'muStruct', muStruct,...
+    'varMean', 0,...
+    'countMseH', 0,...
+    'mseHThreshold', 1.1,...
+    'mseHStoredOld', inf,...
+    'mseHLatestOld', inf,...
+    'delayLatestS', zeros(1,51),...
+    'feedbackDelay', 0,...
+    'feedbackDelayUpdate', 0,...
+    'cntIn', 0,...
+    'cntOut', 0,...
+    'FAR_ENERGY_MIN', 1,...
+    'ENERGY_DEV_OFFSET', 0.5,...
+    'ENERGY_DEV_TOL', 1.5,...
+    'MU_MIN', -16,...
+    'MU_MAX', -2,...
+    'newDelayCurve', 0);
+
+% Adjust speech signals
+xFar = [zeros(setupStruct.hsupport1-1,1);xFar(1:setupStruct.tlength)];
+yNear = [zeros(setupStruct.hsupport1-1,1);yNear(1:setupStruct.tlength)];
+yNearSpeech = [zeros(setupStruct.hsupport1-1,1);yNearSpeech(1:setupStruct.tlength)];
+xFar = xFar(1:setupStruct.tlength);
+yNear = yNear(1:setupStruct.tlength);
+
+% Set figure settings
+if aecmStruct.plotIt
+    figure(13)
+    set(gcf,'doublebuffer','on')
+end
+%%%%%%%%%%
+% Here starts the algorithm
+% Dividing into frames and then estimating the near end speech
+%%%%%%%%%%
+fTheFarEnd      = complex(zeros(setupStruct.hsupport1,1));
+afTheFarEnd     = zeros(setupStruct.hsupport1,setupStruct.updateno+1);
+fFar            = zeros(setupStruct.hsupport1,setupStruct.updateno+1);
+fmicrophone     = complex(zeros(setupStruct.hsupport1,1));
+afmicrophone    = zeros(setupStruct.hsupport1,setupStruct.updateno+1);
+fNear           = zeros(setupStruct.hsupport1,setupStruct.updateno+1);
+femicrophone    = complex(zeros(setupStruct.hsupport1,1));
+emicrophone     = zeros(setupStruct.tlength,1);
+
+if (setupStruct.useDelayEstimation == 2)
+    delSamples = [1641 1895 2032 1895 2311 2000 2350 2222 NaN 2332 2330 2290 2401 2415 NaN 2393 2305 2381 2398];
+    delBlocks = round(delSamples/setupStruct.updatel);
+    delStarts = floor([25138 46844 105991 169901 195739 218536 241803 333905 347703 362660 373753 745135 765887 788078 806257 823835 842443 860139 881869]/setupStruct.updatel);
+else
+    delStarts = [];
+end
+
+for i=1:setupStruct.updateno
+    setupStruct.currentBlock = i;
+    
+    sb = (i-1)*setupStruct.updatel + 1;
+    se = sb + setupStruct.support - 1;
+    
+    %%%%%%%
+    % Analysis FFTs
+    %%%%%%%
+    % Far end signal
+    temp = fft(setupStruct.win .* xFar(sb:se))/frameSize;
+    fTheFarEnd = temp(1:setupStruct.hsupport1);
+    afTheFarEnd(:,i) = abs(fTheFarEnd);
+    fFar(:,i) = fTheFarEnd;
+    % Near end signal
+    temp = fft(setupStruct.win .* yNear(sb:se))/frameSize;%,pause
+    fmicrophone = temp(1:setupStruct.hsupport1);
+    afmicrophone(:,i) = abs(fmicrophone);
+    fNear(:,i) = fmicrophone;
+    %abs(fmicrophone),pause
+    % The true near end speaker (if we have such info)
+    temp = fft(setupStruct.win .* yNearSpeech(sb:se));
+    aftrueSpeech = abs(temp(1:setupStruct.hsupport1));
+    
+    if(i == 1000)
+        %break;
+    end
+    
+    % Perform delay estimation
+    if (setupStruct.useDelayEstimation == 1)
+        % Delay Estimation
+        delayStruct = align(fTheFarEnd, fmicrophone, delayStruct, i);
+        %delayStruct.delay(i) = 39;%19;
+        idel =  max(i - delayStruct.delay(i) + 1,1);
+        
+        if delayCompensation_flag
+            % If we have a new delay estimate from Bastiaan's alg. update the offset
+            if (delayStruct.delay(i) ~= delayStruct.delay(max(1,i-1)))
+                delayStruct.delayAdjust = delayStruct.delayAdjust + delayStruct.delay(i) - delayStruct.delay(i-1);
+            end
+            % Store the compensated delay
+            delayStruct.delayNew(i) = delayStruct.delay(i) - delayStruct.delayAdjust;
+            if (delayStruct.delayNew(i) < 1)
+                % Something's wrong
+                pause,break
+            end
+            % Compensate with the offset estimate
+            idel = idel + delayStruct.delayAdjust;
+        end
+        if 0%aecmStruct.plotIt
+            figure(1)
+            plot(1:i,delayStruct.delay(1:i),'k:',1:i,delayStruct.delayNew(1:i),'k--','LineWidth',2),drawnow
+        end
+    elseif (setupStruct.useDelayEstimation == 2)
+        % Use "manual delay"
+        delIndex = find(delStarts<i);
+        if isempty(delIndex)
+            idel = i;
+        else
+            idel = i - delBlocks(max(delIndex));
+            if isnan(idel)
+                idel = i - delBlocks(max(delIndex)-1);
+            end
+        end
+    else
+        % No delay estimation
+        %idel = max(i - 18, 1);
+        idel = max(i - 50, 1);
+    end
+
+    %%%%%%%%
+    % This is the AECM algorithm
+    %
+    % Output is the new frequency domain signal (hopefully) echo compensated
+    %%%%%%%%
+    [femicrophone, aecmStruct] = AECMobile(fmicrophone, afTheFarEnd(:,idel), setupStruct, aecmStruct);
+    %[femicrophone, aecmStruct] = AECMobile(fmicrophone, FARENDFFT(idel,:)'/2^F(idel,end-1), setupStruct, aecmStruct);
+    
+    if aecmStruct.feedbackDelayUpdate
+        % If the feedback tells us there is a new offset out there update the enhancement
+        delayStruct.delayAdjust = delayStruct.delayAdjust + aecmStruct.feedbackDelay;
+        aecmStruct.feedbackDelayUpdate = 0;
+    end
+    
+    % reconstruction; first make spectrum odd
+    temp = [femicrophone; flipud(conj(femicrophone(2:(setupStruct.hsupport1-1))))];
+    emicrophone(sb:se) = emicrophone(sb:se) + setupStruct.factor * setupStruct.win .* real(ifft(temp))*frameSize;
+    if max(isnan(emicrophone(sb:se)))
+        % Something's wrong with the output at block i
+        i
+        break
+    end
+end
+
+
+if useHTC
+    fid=fopen('aecOutMatlabC.pcm','w');fwrite(fid,int16(emicrophone),'short');fclose(fid);
+    %fid=fopen('farendFFT.txt','w');fwrite(fid,int16(afTheFarEnd(:)),'short');fclose(fid);
+    %fid=fopen('farendFFTreal.txt','w');fwrite(fid,int16(imag(fFar(:))),'short');fclose(fid);
+    %fid=fopen('farendFFTimag.txt','w');fwrite(fid,int16(real(fFar(:))),'short');fclose(fid);
+    %fid=fopen('nearendFFT.txt','w');fwrite(fid,int16(afmicrophone(:)),'short');fclose(fid);
+    %fid=fopen('nearendFFTreal.txt','w');fwrite(fid,int16(real(fNear(:))),'short');fclose(fid);
+    %fid=fopen('nearendFFTimag.txt','w');fwrite(fid,int16(imag(fNear(:))),'short');fclose(fid);
+end
+if useHTC
+    %spclab(setupStruct.samplingfreq,xFar,yNear,emicrophone)
+else
+    spclab(setupStruct.samplingfreq,xFar,yNear,emicrophone,yNearSpeech)
+end    

src/modules/audio_processing/aecm/main/matlab/matlab/simEnvironment.m

@@ -0,0 +1,15 @@
+speakerType = 'fm';
+%for k=2:5
+%for k=[2 4 5]
+for k=3
+    scenario = int2str(k);
+    fprintf('Current scenario: %d\n',k)
+    mainProgram
+    %saveFile = [speakerType, '_s_',scenario,'_delayEst_v2_vad_man.wav'];
+    %wavwrite(emic,fs,nbits,saveFile);
+    %saveFile = ['P:\Engineering_share\BjornV\AECM\',speakerType, '_s_',scenario,'_delayEst_v2_vad_man.pcm'];
+    %saveFile = [speakerType, '_s_',scenario,'_adaptMu_adaptGamma_withVar_gammFilt_HSt.pcm'];
+    saveFile = ['scenario_',scenario,'_090417_backupH_nlp.pcm'];
+    fid=fopen(saveFile,'w');fwrite(fid,int16(emicrophone),'short');fclose(fid);
+    %pause
+end

src/modules/audio_processing/aecm/main/matlab/matlab/updateSettings.m

@@ -0,0 +1,94 @@
+function [setupStructNew, delayStructNew] = updateSettings(microphone, TheFarEnd, setupStruct, delayStruct);
+
+% other, constants
+setupStruct.hsupport1 = setupStruct.support/2 + 1;
+setupStruct.factor =  2 / setupStruct.oversampling;
+setupStruct.updatel = setupStruct.support/setupStruct.oversampling;
+setupStruct.estLen = round(setupStruct.avtime * setupStruct.samplingfreq/setupStruct.updatel);
+
+% compute some constants
+blockLen = setupStruct.support/setupStruct.oversampling;
+delayStruct.maxDelayb = floor(setupStruct.samplingfreq*delayStruct.maxDelay/setupStruct.updatel); % in blocks
+
+%input
+tlength = min([length(microphone),length(TheFarEnd)]);
+updateno = floor(tlength/setupStruct.updatel);
+setupStruct.tlength = setupStruct.updatel*updateno;
+setupStruct.updateno = updateno - setupStruct.oversampling + 1;
+
+% signal length
+n = floor(min([length(TheFarEnd), length(microphone)])/setupStruct.support)*setupStruct.support;
+setupStruct.nb = n/blockLen - setupStruct.oversampling + 1; % in blocks
+
+setupStruct.win = sqrt([0 ; hanning(setupStruct.support-1)]);
+
+% Construct filterbank in Bark-scale
+
+K = setupStruct.subBandLength; %Something's wrong when K even
+erbs = 21.4*log10(0.00437*setupStruct.samplingfreq/2+1);
+fe = (10.^((0:K)'*erbs/K/21.4)-1)/0.00437;
+setupStruct.centerFreq = fe;
+H = diag(ones(1,K-1))+diag(ones(1,K-2),-1);
+Hinv = inv(H);
+aty = 2*Hinv(end,:)*fe(2:end-1);
+boundary = aty - (setupStruct.samplingfreq/2 + fe(end-1))/2;
+if rem(K,2)
+    x1 = min([fe(2)/2, -boundary]);
+else
+    x1 = max([0, boundary]);
+end
+%x1
+g = fe(2:end-1);
+g(1) = g(1) - x1/2;
+x = 2*Hinv*g;
+x = [x1;x];
+%figure(42), clf
+xy = zeros((K+1)*4,1);
+yy = zeros((K+1)*4,1);
+xy(1:4) = [fe(1) fe(1) x(1) x(1)]';
+yy(1:4) = [0 1 1 0]'/x(1);
+for kk=2:K
+    xy((kk-1)*4+(1:4)) = [x(kk-1) x(kk-1) x(kk) x(kk)]';
+    yy((kk-1)*4+(1:4)) = [0 1 1 0]'/(x(kk)-x(kk-1));
+end
+xy(end-3:end) = [x(K) x(K) fe(end) fe(end)]';
+yy(end-3:end) = [0 1 1 0]'/(fe(end)*2-2*x(K));
+%plot(xy,yy,'LineWidth',2)
+%fill(xy,yy,'y')
+
+x = [0;x];
+xk = x*setupStruct.hsupport1/setupStruct.samplingfreq*2;
+%setupStruct.erbBoundaries = xk;
+numInBand = zeros(length(xk),1);
+xh = (0:setupStruct.hsupport1-1);
+
+for kk=1:length(xk)
+    if (kk==length(xk))
+        numInBand(kk) = length(find(xh>=xk(kk)));
+    else
+        numInBand(kk) = length(intersect(find(xh>=xk(kk)),find(xh<xk(kk+1))));
+    end
+end
+setupStruct.numInBand = numInBand;
+
+setupStructNew = setupStruct;
+
+delayStructNew = struct(...
+    'sxAll2',zeros(setupStructNew.hsupport1,setupStructNew.nb),...
+    'syAll2',zeros(setupStructNew.hsupport1,setupStructNew.nb),...
+    'z200',zeros(5,setupStructNew.hsupport1),...
+    'z500',zeros(5,delayStruct.maxDelayb+1),...
+    'bxspectrum',uint32(zeros(setupStructNew.nb,1)),...
+    'byspectrum',uint32(zeros(setupStructNew.nb,1)),...
+    'bandfirst',delayStruct.bandfirst,'bandlast',delayStruct.bandlast,...
+    'bxhist',uint32(zeros(delayStruct.maxDelayb+1,1)),...
+    'bcount',zeros(1+delayStruct.maxDelayb,setupStructNew.nb),...
+    'fout',zeros(1+delayStruct.maxDelayb,setupStructNew.nb),...
+    'new',zeros(1+delayStruct.maxDelayb,setupStructNew.nb),...
+    'smlength',delayStruct.smlength,...
+    'maxDelay', delayStruct.maxDelay,...
+    'maxDelayb', delayStruct.maxDelayb,...
+    'oneGoodEstimate', 0,...
+    'delayAdjust', 0,...
+    'delayNew',zeros(setupStructNew.nb,1),...
+    'delay',zeros(setupStructNew.nb,1));

src/modules/audio_processing/aecm/main/matlab/waitbar_j.m

@@ -0,0 +1,234 @@
+function fout = waitbar_j(x,whichbar, varargin)
+%WAITBAR Display wait bar.
+%   H = WAITBAR(X,'title', property, value, property, value, ...) 
+%   creates and displays a waitbar of fractional length X.  The 
+%   handle to the waitbar figure is returned in H.
+%   X should be between 0 and 1.  Optional arguments property and 
+%   value allow to set corresponding waitbar figure properties.
+%   Property can also be an action keyword 'CreateCancelBtn', in 
+%   which case a cancel button will be added to the figure, and 
+%   the passed value string will be executed upon clicking on the 
+%   cancel button or the close figure button.
+%
+%   WAITBAR(X) will set the length of the bar in the most recently
+%   created waitbar window to the fractional length X.
+%
+%   WAITBAR(X,H) will set the length of the bar in waitbar H
+%   to the fractional length X.
+%
+%   WAITBAR(X,H,'updated title') will update the title text in
+%   the waitbar figure, in addition to setting the fractional
+%   length to X.
+%
+%   WAITBAR is typically used inside a FOR loop that performs a 
+%   lengthy computation.  A sample usage is shown below:
+%
+%       h = waitbar(0,'Please wait...');
+%       for i=1:100,
+%           % computation here %
+%           waitbar(i/100,h)
+%       end
+%       close(h)
+
+%   Clay M. Thompson 11-9-92
+%   Vlad Kolesnikov  06-7-99
+%   Copyright 1984-2001 The MathWorks, Inc.
+%   $Revision: 1.22 $  $Date: 2001/04/15 12:03:29 $
+
+if nargin>=2
+    if ischar(whichbar)
+        type=2; %we are initializing
+        name=whichbar;
+    elseif isnumeric(whichbar)
+        type=1; %we are updating, given a handle
+        f=whichbar;
+    else
+        error(['Input arguments of type ' class(whichbar) ' not valid.'])
+    end
+elseif nargin==1
+    f = findobj(allchild(0),'flat','Tag','TMWWaitbar');
+    
+    if isempty(f)
+        type=2;
+        name='Waitbar';
+    else
+        type=1;
+        f=f(1);
+    end   
+else
+    error('Input arguments not valid.');
+end
+
+x = max(0,min(100*x,100));
+
+switch type
+ case 1,  % waitbar(x)    update
+  p = findobj(f,'Type','patch');
+  l = findobj(f,'Type','line');
+  if isempty(f) | isempty(p) | isempty(l), 
+      error('Couldn''t find waitbar handles.'); 
+  end
+  xpatch = get(p,'XData');
+  xpatch = [0 x x 0];
+  set(p,'XData',xpatch)
+  xline = get(l,'XData');
+  set(l,'XData',xline);
+  
+  if nargin>2,
+      % Update waitbar title:
+      hAxes = findobj(f,'type','axes');
+      hTitle = get(hAxes,'title');
+      set(hTitle,'string',varargin{1});
+  end
+  
+ case 2,  % waitbar(x,name)  initialize
+  vertMargin = 0;
+  if nargin > 2,
+      % we have optional arguments: property-value pairs
+      if rem (nargin, 2 ) ~= 0
+          error( 'Optional initialization arguments must be passed in pairs' );
+      end
+  end
+  
+  oldRootUnits = get(0,'Units');
+
+  set(0, 'Units', 'points');
+  screenSize = get(0,'ScreenSize');
+  
+  axFontSize=get(0,'FactoryAxesFontSize');
+  
+  pointsPerPixel = 72/get(0,'ScreenPixelsPerInch');
+  
+  width = 360 * pointsPerPixel;
+  height = 75 * pointsPerPixel;
+  pos = [screenSize(3)/2-width/2 screenSize(4)/2-height/2 width height];
+
+%pos=  [501.75 589.5 393.75 52.5];
+  f = figure(...
+      'Units', 'points', ...
+      'BusyAction', 'queue', ...
+      'Position', pos, ...
+      'Resize','on', ...
+      'CreateFcn','', ...
+      'NumberTitle','off', ...
+      'IntegerHandle','off', ...
+      'MenuBar', 'none', ...
+      'Tag','TMWWaitbar',...
+      'Interruptible', 'off', ...
+      'Visible','on');
+  
+  %%%%%%%%%%%%%%%%%%%%%
+  % set figure properties as passed to the fcn
+  % pay special attention to the 'cancel' request
+  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
+  if nargin > 2,
+      propList = varargin(1:2:end);
+      valueList = varargin(2:2:end);
+      cancelBtnCreated = 0;
+      for ii = 1:length( propList )
+          try
+              if strcmp(lower(propList{ii}), 'createcancelbtn' ) & ~cancelBtnCreated
+                  cancelBtnHeight = 23 * pointsPerPixel;
+                  cancelBtnWidth = 60 * pointsPerPixel;
+                  newPos = pos;
+                  vertMargin = vertMargin + cancelBtnHeight;
+                  newPos(4) = newPos(4)+vertMargin;
+                  callbackFcn = [valueList{ii}];
+                  set( f, 'Position', newPos, 'CloseRequestFcn', callbackFcn );
+                  cancelButt = uicontrol('Parent',f, ...
+                                         'Units','points', ...
+                                         'Callback',callbackFcn, ...
+                                         'ButtonDownFcn', callbackFcn, ...
+                                         'Enable','on', ...
+                                         'Interruptible','off', ...
+                                         'Position', [pos(3)-cancelBtnWidth*1.4, 7,  ...
+                    cancelBtnWidth, cancelBtnHeight], ...
+                                         'String','Cancel', ...
+                                         'Tag','TMWWaitbarCancelButton');
+                  cancelBtnCreated = 1;
+              else
+                  % simply set the prop/value pair of the figure
+                  set( f, propList{ii}, valueList{ii});
+              end
+          catch
+              disp ( ['Warning: could not set property ''' propList{ii} ''' with value ''' num2str(valueList{ii}) '''' ] );
+          end
+      end
+  end  
+  
+  %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%  
+  
+  
+  colormap([]);
+  
+  axNorm=[.05 .3 .9 .2];
+ % axNorm=[1 1 1 1];
+  axPos=axNorm.*[pos(3:4),pos(3:4)] + [0 vertMargin 0 0];
+  
+  h = axes('XLim',[0 100],...
+           'YLim',[0 1],...
+           'Box','on', ...
+           'Units','Points',...
+           'FontSize', axFontSize,...
+           'Position',axPos,...
+           'XTickMode','manual',...
+           'YTickMode','manual',...
+           'XTick',[],...
+           'YTick',[],...
+           'XTickLabelMode','manual',...
+           'XTickLabel',[],...
+           'YTickLabelMode','manual',...
+           'YTickLabel',[]);
+  
+  tHandle=title(name);
+  tHandle=get(h,'title');
+  oldTitleUnits=get(tHandle,'Units');
+  set(tHandle,...
+      'Units',      'points',...
+      'String',     name);
+  
+  tExtent=get(tHandle,'Extent');
+  set(tHandle,'Units',oldTitleUnits);
+  
+  titleHeight=tExtent(4)+axPos(2)+axPos(4)+5;
+  if titleHeight>pos(4)
+      pos(4)=titleHeight;
+      pos(2)=screenSize(4)/2-pos(4)/2;
+      figPosDirty=logical(1);
+  else
+      figPosDirty=logical(0);
+  end
+  
+  if tExtent(3)>pos(3)*1.10;
+      pos(3)=min(tExtent(3)*1.10,screenSize(3));
+      pos(1)=screenSize(3)/2-pos(3)/2;
+      
+      axPos([1,3])=axNorm([1,3])*pos(3);
+      set(h,'Position',axPos);
+      
+      figPosDirty=logical(1);
+  end
+  
+  if figPosDirty
+      set(f,'Position',pos);
+  end
+
+  xpatch = [0 x x 0];
+  ypatch = [0 0 1 1];
+   xline = [100 0 0 100 100];
+   yline = [0 0 1 1 0];
+  
+  p = patch(xpatch,ypatch,'r','EdgeColor','r','EraseMode','none');
+  l = line(xline,yline,'EraseMode','none');
+  set(l,'Color',get(gca,'XColor'));
+  
+  
+  set(f,'HandleVisibility','callback','visible','on', 'resize','off');
+  
+  set(0, 'Units', oldRootUnits);
+end  % case
+drawnow;
+
+if nargout==1,
+    fout = f;
+end

src/modules/audio_processing/aecm/main/source/aecm.gypi

@@ -0,0 +1,42 @@
+# Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+#
+# Use of this source code is governed by a BSD-style license
+# that can be found in the LICENSE file in the root of the source
+# tree. An additional intellectual property rights grant can be found
+# in the file PATENTS.  All contributing project authors may
+# be found in the AUTHORS file in the root of the source tree.
+
+{
+  'targets': [
+    {
+      'target_name': 'aecm',
+      'type': '<(library)',
+      'dependencies': [
+        '<(webrtc_root)/common_audio/common_audio.gyp:spl',
+        'apm_util'
+      ],
+      'include_dirs': [
+        '../interface',
+      ],
+      'direct_dependent_settings': {
+        'include_dirs': [
+          '../interface',
+        ],
+      },
+      'sources': [
+        '../interface/echo_control_mobile.h',
+        'echo_control_mobile.c',
+        'aecm_core.c',
+        'aecm_core.h',
+        'aecm_delay_estimator.c',
+        'aecm_delay_estimator.h',
+      ],
+    },
+  ],
+}
+
+# Local Variables:
+# tab-width:2
+# indent-tabs-mode:nil
+# End:
+# vim: set expandtab tabstop=2 shiftwidth=2:

src/modules/audio_processing/aecm/main/source/aecm_core.h

@@ -0,0 +1,358 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+// Performs echo control (suppression) with fft routines in fixed-point
+
+#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AECM_MAIN_SOURCE_AECM_CORE_H_
+#define WEBRTC_MODULES_AUDIO_PROCESSING_AECM_MAIN_SOURCE_AECM_CORE_H_
+
+#define AECM_DYNAMIC_Q // turn on/off dynamic Q-domain
+//#define AECM_WITH_ABS_APPROX
+//#define AECM_SHORT                // for 32 sample partition length (otherwise 64)
+
+#include "typedefs.h"
+#include "signal_processing_library.h"
+
+// Algorithm parameters
+
+#define FRAME_LEN       80              // Total frame length, 10 ms
+#ifdef AECM_SHORT
+
+#define PART_LEN        32              // Length of partition
+#define PART_LEN_SHIFT  6               // Length of (PART_LEN * 2) in base 2
+
+#else
+
+#define PART_LEN        64              // Length of partition
+#define PART_LEN_SHIFT  7               // Length of (PART_LEN * 2) in base 2
+
+#endif
+
+#define PART_LEN1       (PART_LEN + 1)  // Unique fft coefficients
+#define PART_LEN2       (PART_LEN << 1) // Length of partition * 2
+#define PART_LEN4       (PART_LEN << 2) // Length of partition * 4
+#define FAR_BUF_LEN     PART_LEN4       // Length of buffers
+#define MAX_DELAY 100
+
+// Counter parameters
+#ifdef AECM_SHORT
+
+#define CONV_LEN        1024            // Convergence length used at startup
+#else
+
+#define CONV_LEN        512             // Convergence length used at startup
+#endif
+
+#define CONV_LEN2       (CONV_LEN << 1) // Convergence length * 2 used at startup
+// Energy parameters
+#define MAX_BUF_LEN     64              // History length of energy signals
+
+#define FAR_ENERGY_MIN  1025            // Lowest Far energy level: At least 2 in energy
+#define FAR_ENERGY_DIFF 929             // Allowed difference between max and min
+
+#define ENERGY_DEV_OFFSET       0       // The energy error offset in Q8
+#define ENERGY_DEV_TOL  400             // The energy estimation tolerance in Q8
+#define FAR_ENERGY_VAD_REGION   230     // Far VAD tolerance region
+// Stepsize parameters
+#define MU_MIN          10              // Min stepsize 2^-MU_MIN (far end energy dependent)
+#define MU_MAX          1               // Max stepsize 2^-MU_MAX (far end energy dependent)
+#define MU_DIFF         9               // MU_MIN - MU_MAX
+// Channel parameters
+#define MIN_MSE_COUNT   20              // Min number of consecutive blocks with enough far end
+                                        // energy to compare channel estimates
+#define MIN_MSE_DIFF    29              // The ratio between adapted and stored channel to
+                                        // accept a new storage (0.8 in Q-MSE_RESOLUTION)
+#define MSE_RESOLUTION  5               // MSE parameter resolution
+#define RESOLUTION_CHANNEL16    12      // W16 Channel in Q-RESOLUTION_CHANNEL16
+#define RESOLUTION_CHANNEL32    28      // W32 Channel in Q-RESOLUTION_CHANNEL
+#define CHANNEL_VAD     16              // Minimum energy in frequency band to update channel
+// Suppression gain parameters: SUPGAIN_ parameters in Q-(RESOLUTION_SUPGAIN)
+#define RESOLUTION_SUPGAIN      8       // Channel in Q-(RESOLUTION_SUPGAIN)
+#define SUPGAIN_DEFAULT (1 << RESOLUTION_SUPGAIN)   // Default suppression gain
+#define SUPGAIN_ERROR_PARAM_A   3072    // Estimation error parameter (Maximum gain) (8 in Q8)
+#define SUPGAIN_ERROR_PARAM_B   1536    // Estimation error parameter (Gain before going down)
+#define SUPGAIN_ERROR_PARAM_D   SUPGAIN_DEFAULT // Estimation error parameter
+                                                // (Should be the same as Default) (1 in Q8)
+#define SUPGAIN_EPC_DT  200             // = SUPGAIN_ERROR_PARAM_C * ENERGY_DEV_TOL
+// Defines for "check delay estimation"
+#define CORR_WIDTH      31              // Number of samples to correlate over.
+#define CORR_MAX        16              // Maximum correlation offset
+#define CORR_MAX_BUF    63
+#define CORR_DEV        4
+#define CORR_MAX_LEVEL  20
+#define CORR_MAX_LOW    4
+#define CORR_BUF_LEN    (CORR_MAX << 1) + 1
+// Note that CORR_WIDTH + 2*CORR_MAX <= MAX_BUF_LEN
+
+#define ONE_Q14         (1 << 14)
+
+// NLP defines
+#define NLP_COMP_LOW    3277            // 0.2 in Q14
+#define NLP_COMP_HIGH   ONE_Q14         // 1 in Q14
+
+extern const WebRtc_Word16 WebRtcAecm_kSqrtHanning[];
+
+typedef struct {
+    WebRtc_Word16 real;
+    WebRtc_Word16 imag;
+} complex16_t;
+
+typedef struct
+{
+    int farBufWritePos;
+    int farBufReadPos;
+    int knownDelay;
+    int lastKnownDelay;
+    int firstVAD; // Parameter to control poorly initialized channels
+
+    void *farFrameBuf;
+    void *nearNoisyFrameBuf;
+    void *nearCleanFrameBuf;
+    void *outFrameBuf;
+
+    WebRtc_Word16 farBuf[FAR_BUF_LEN];
+
+    WebRtc_Word16 mult;
+    WebRtc_UWord32 seed;
+
+    // Delay estimation variables
+    void* delay_estimator;
+    WebRtc_UWord16 currentDelay;
+
+    WebRtc_Word16 nlpFlag;
+    WebRtc_Word16 fixedDelay;
+
+    WebRtc_UWord32 totCount;
+
+    WebRtc_Word16 dfaCleanQDomain;
+    WebRtc_Word16 dfaCleanQDomainOld;
+    WebRtc_Word16 dfaNoisyQDomain;
+    WebRtc_Word16 dfaNoisyQDomainOld;
+
+    WebRtc_Word16 nearLogEnergy[MAX_BUF_LEN];
+    WebRtc_Word16 farLogEnergy;
+    WebRtc_Word16 echoAdaptLogEnergy[MAX_BUF_LEN];
+    WebRtc_Word16 echoStoredLogEnergy[MAX_BUF_LEN];
+
+    // The extra 16 or 32 bytes in the following buffers are for alignment based Neon code.
+    // It's designed this way since the current GCC compiler can't align a buffer in 16 or 32
+    // byte boundaries properly.
+    WebRtc_Word16 channelStored_buf[PART_LEN1 + 8];
+    WebRtc_Word16 channelAdapt16_buf[PART_LEN1 + 8];
+    WebRtc_Word32 channelAdapt32_buf[PART_LEN1 + 8];
+    WebRtc_Word16 xBuf_buf[PART_LEN2 + 16]; // farend
+    WebRtc_Word16 dBufClean_buf[PART_LEN2 + 16]; // nearend
+    WebRtc_Word16 dBufNoisy_buf[PART_LEN2 + 16]; // nearend
+    WebRtc_Word16 outBuf_buf[PART_LEN + 8];
+
+    // Pointers to the above buffers
+    WebRtc_Word16 *channelStored;
+    WebRtc_Word16 *channelAdapt16;
+    WebRtc_Word32 *channelAdapt32;
+    WebRtc_Word16 *xBuf;
+    WebRtc_Word16 *dBufClean;
+    WebRtc_Word16 *dBufNoisy;
+    WebRtc_Word16 *outBuf;
+
+    WebRtc_Word32 echoFilt[PART_LEN1];
+    WebRtc_Word16 nearFilt[PART_LEN1];
+    WebRtc_Word32 noiseEst[PART_LEN1];
+    int           noiseEstTooLowCtr[PART_LEN1];
+    int           noiseEstTooHighCtr[PART_LEN1];
+    WebRtc_Word16 noiseEstCtr;
+    WebRtc_Word16 cngMode;
+
+    WebRtc_Word32 mseAdaptOld;
+    WebRtc_Word32 mseStoredOld;
+    WebRtc_Word32 mseThreshold;
+
+    WebRtc_Word16 farEnergyMin;
+    WebRtc_Word16 farEnergyMax;
+    WebRtc_Word16 farEnergyMaxMin;
+    WebRtc_Word16 farEnergyVAD;
+    WebRtc_Word16 farEnergyMSE;
+    WebRtc_Word16 currentVADValue;
+    WebRtc_Word16 vadUpdateCount;
+
+    WebRtc_Word16 startupState;
+    WebRtc_Word16 mseChannelCount;
+    WebRtc_Word16 supGain;
+    WebRtc_Word16 supGainOld;
+
+    WebRtc_Word16 supGainErrParamA;
+    WebRtc_Word16 supGainErrParamD;
+    WebRtc_Word16 supGainErrParamDiffAB;
+    WebRtc_Word16 supGainErrParamDiffBD;
+
+#ifdef AEC_DEBUG
+    FILE *farFile;
+    FILE *nearFile;
+    FILE *outFile;
+#endif
+} AecmCore_t;
+
+///////////////////////////////////////////////////////////////////////////////////////////////
+// WebRtcAecm_CreateCore(...)
+//
+// Allocates the memory needed by the AECM. The memory needs to be
+// initialized separately using the WebRtcAecm_InitCore() function.
+//
+// Input:
+//      - aecm          : Instance that should be created
+//
+// Output:
+//      - aecm          : Created instance
+//
+// Return value         :  0 - Ok
+//                        -1 - Error
+//
+int WebRtcAecm_CreateCore(AecmCore_t **aecm);
+
+///////////////////////////////////////////////////////////////////////////////////////////////
+// WebRtcAecm_InitCore(...)
+//
+// This function initializes the AECM instant created with WebRtcAecm_CreateCore(...)
+// Input:
+//      - aecm          : Pointer to the AECM instance
+//      - samplingFreq  : Sampling Frequency
+//
+// Output:
+//      - aecm          : Initialized instance
+//
+// Return value         :  0 - Ok
+//                        -1 - Error
+//
+int WebRtcAecm_InitCore(AecmCore_t * const aecm, int samplingFreq);
+
+///////////////////////////////////////////////////////////////////////////////////////////////
+// WebRtcAecm_FreeCore(...)
+//
+// This function releases the memory allocated by WebRtcAecm_CreateCore()
+// Input:
+//      - aecm          : Pointer to the AECM instance
+//
+// Return value         :  0 - Ok
+//                        -1 - Error
+//           11001-11016: Error
+//
+int WebRtcAecm_FreeCore(AecmCore_t *aecm);
+
+int WebRtcAecm_Control(AecmCore_t *aecm, int delay, int nlpFlag);
+
+///////////////////////////////////////////////////////////////////////////////////////////////
+// WebRtcAecm_InitEchoPathCore(...)
+//
+// This function resets the echo channel adaptation with the specified channel.
+// Input:
+//      - aecm          : Pointer to the AECM instance
+//      - echo_path     : Pointer to the data that should initialize the echo path
+//
+// Output:
+//      - aecm          : Initialized instance
+//
+void WebRtcAecm_InitEchoPathCore(AecmCore_t* aecm, const WebRtc_Word16* echo_path);
+
+///////////////////////////////////////////////////////////////////////////////////////////////
+// WebRtcAecm_ProcessFrame(...)
+//
+// This function processes frames and sends blocks to WebRtcAecm_ProcessBlock(...)
+//
+// Inputs:
+//      - aecm          : Pointer to the AECM instance
+//      - farend        : In buffer containing one frame of echo signal
+//      - nearendNoisy  : In buffer containing one frame of nearend+echo signal without NS
+//      - nearendClean  : In buffer containing one frame of nearend+echo signal with NS
+//
+// Output:
+//      - out           : Out buffer, one frame of nearend signal          :
+//
+//
+int WebRtcAecm_ProcessFrame(AecmCore_t * aecm, const WebRtc_Word16 * farend,
+                            const WebRtc_Word16 * nearendNoisy,
+                            const WebRtc_Word16 * nearendClean,
+                            WebRtc_Word16 * out);
+
+///////////////////////////////////////////////////////////////////////////////////////////////
+// WebRtcAecm_ProcessBlock(...)
+//
+// This function is called for every block within one frame
+// This function is called by WebRtcAecm_ProcessFrame(...)
+//
+// Inputs:
+//      - aecm          : Pointer to the AECM instance
+//      - farend        : In buffer containing one block of echo signal
+//      - nearendNoisy  : In buffer containing one frame of nearend+echo signal without NS
+//      - nearendClean  : In buffer containing one frame of nearend+echo signal with NS
+//
+// Output:
+//      - out           : Out buffer, one block of nearend signal          :
+//
+//
+int WebRtcAecm_ProcessBlock(AecmCore_t * aecm, const WebRtc_Word16 * farend,
+                            const WebRtc_Word16 * nearendNoisy,
+                            const WebRtc_Word16 * noisyClean,
+                            WebRtc_Word16 * out);
+
+///////////////////////////////////////////////////////////////////////////////////////////////
+// WebRtcAecm_BufferFarFrame()
+//
+// Inserts a frame of data into farend buffer.
+//
+// Inputs:
+//      - aecm          : Pointer to the AECM instance
+//      - farend        : In buffer containing one frame of farend signal
+//      - farLen        : Length of frame
+//
+void WebRtcAecm_BufferFarFrame(AecmCore_t * const aecm, const WebRtc_Word16 * const farend,
+                               const int farLen);
+
+///////////////////////////////////////////////////////////////////////////////////////////////
+// WebRtcAecm_FetchFarFrame()
+//
+// Read the farend buffer to account for known delay
+//
+// Inputs:
+//      - aecm          : Pointer to the AECM instance
+//      - farend        : In buffer containing one frame of farend signal
+//      - farLen        : Length of frame
+//      - knownDelay    : known delay
+//
+void WebRtcAecm_FetchFarFrame(AecmCore_t * const aecm, WebRtc_Word16 * const farend,
+                              const int farLen, const int knownDelay);
+
+///////////////////////////////////////////////////////////////////////////////////////////////
+// Some internal functions shared by ARM NEON and generic C code:
+//
+
+void WebRtcAecm_CalcLinearEnergies(AecmCore_t* aecm,
+                                   const WebRtc_UWord16* far_spectrum,
+                                   WebRtc_Word32* echoEst,
+                                   WebRtc_UWord32* far_energy,
+                                   WebRtc_UWord32* echo_energy_adapt,
+                                   WebRtc_UWord32* echo_energy_stored);
+
+void WebRtcAecm_StoreAdaptiveChannel(AecmCore_t* aecm,
+                                     const WebRtc_UWord16* far_spectrum,
+                                     WebRtc_Word32* echo_est);
+
+void WebRtcAecm_ResetAdaptiveChannel(AecmCore_t *aecm);
+
+void WebRtcAecm_WindowAndFFT(WebRtc_Word16* fft,
+                             const WebRtc_Word16* time_signal,
+                             complex16_t* freq_signal,
+                             int time_signal_scaling);
+
+void WebRtcAecm_InverseFFTAndWindow(AecmCore_t* aecm,
+                                    WebRtc_Word16* fft,
+                                    complex16_t* efw,
+                                    WebRtc_Word16* output,
+                                    const WebRtc_Word16* nearendClean);
+
+#endif

src/modules/audio_processing/aecm/main/source/aecm_core_neon.c

@@ -0,0 +1,314 @@
+/*
+ *  Copyright (c) 2011 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+#if defined(WEBRTC_ANDROID) && defined(WEBRTC_ARCH_ARM_NEON)
+
+#include "aecm_core.h"
+
+#include <arm_neon.h>
+#include <assert.h>
+
+
+// Square root of Hanning window in Q14.
+static const WebRtc_Word16 kSqrtHanningReversed[] __attribute__ ((aligned (8))) = {       
+     16384, 16373, 16354, 16325,  
+     16286, 16237, 16179, 16111,  
+     16034, 15947, 15851, 15746,  
+     15631, 15506, 15373, 15231,  
+     15079, 14918, 14749, 14571,  
+     14384, 14189, 13985, 13773,  
+     13553, 13325, 13089, 12845,  
+     12594, 12335, 12068, 11795,  
+     11514, 11227, 10933, 10633,  
+     10326, 10013, 9695,  9370,   
+     9040,  8705,  8364,  8019,   
+     7668,  7313,  6954,  6591,   
+     6224,  5853,  5478,  5101,   
+     4720,  4337,  3951,  3562,   
+     3172,  2780,  2386,  1990,   
+     1594,  1196,  798,   399
+};
+
+void WebRtcAecm_WindowAndFFT(WebRtc_Word16* fft,
+                             const WebRtc_Word16* time_signal,
+                             complex16_t* freq_signal,
+                             int time_signal_scaling)
+{
+    int i, j;
+
+    int16x4_t tmp16x4_scaling = vdup_n_s16(time_signal_scaling);
+    __asm__("vmov.i16 d21, #0" ::: "d21");
+
+    for(i = 0, j = 0; i < PART_LEN; i += 4, j += 8)
+    {
+        int16x4_t tmp16x4_0;
+        int16x4_t tmp16x4_1;
+        int32x4_t tmp32x4_0;
+
+        /* Window near end */
+        // fft[j] = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT((time_signal[i]
+        //       << time_signal_scaling), WebRtcAecm_kSqrtHanning[i], 14);
+        __asm__("vld1.16 %P0, [%1, :64]" : "=w"(tmp16x4_0) : "r"(&time_signal[i]));
+        tmp16x4_0 = vshl_s16(tmp16x4_0, tmp16x4_scaling);
+
+        __asm__("vld1.16 %P0, [%1, :64]" : "=w"(tmp16x4_1) : "r"(&WebRtcAecm_kSqrtHanning[i]));
+        tmp32x4_0 = vmull_s16(tmp16x4_0, tmp16x4_1);
+
+        __asm__("vshrn.i32 d20, %q0, #14" : : "w"(tmp32x4_0) : "d20");
+        __asm__("vst2.16 {d20, d21}, [%0, :128]" : : "r"(&fft[j]) : "q10");
+
+        // fft[PART_LEN2 + j] = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT(
+        //      (time_signal[PART_LEN + i] << time_signal_scaling),
+        //       WebRtcAecm_kSqrtHanning[PART_LEN - i], 14);
+        __asm__("vld1.16 %P0, [%1, :64]" : "=w"(tmp16x4_0) : "r"(&time_signal[i + PART_LEN]));
+        tmp16x4_0 = vshl_s16(tmp16x4_0, tmp16x4_scaling);
+
+        __asm__("vld1.16 %P0, [%1, :64]" : "=w"(tmp16x4_1) : "r"(&kSqrtHanningReversed[i]));
+        tmp32x4_0 = vmull_s16(tmp16x4_0, tmp16x4_1);
+
+        __asm__("vshrn.i32 d20, %q0, #14" : : "w"(tmp32x4_0) : "d20");
+        __asm__("vst2.16 {d20, d21}, [%0, :128]" : : "r"(&fft[PART_LEN2 + j]) : "q10");
+    }
+
+    WebRtcSpl_ComplexBitReverse(fft, PART_LEN_SHIFT);
+    WebRtcSpl_ComplexFFT(fft, PART_LEN_SHIFT, 1);
+
+    // Take only the first PART_LEN2 samples, and switch the sign of the imaginary part.
+    for(i = 0, j = 0; j < PART_LEN2; i += 8, j += 16)
+    {
+        __asm__("vld2.16 {d20, d21, d22, d23}, [%0, :256]" : : "r"(&fft[j]) : "q10", "q11");
+        __asm__("vneg.s16 d22, d22" : : : "q10");
+        __asm__("vneg.s16 d23, d23" : : : "q11");
+        __asm__("vst2.16 {d20, d21, d22, d23}, [%0, :256]" : : 
+            "r"(&freq_signal[i].real): "q10", "q11");
+    }
+}
+
+void WebRtcAecm_InverseFFTAndWindow(AecmCore_t* aecm,
+                        WebRtc_Word16* fft,
+                        complex16_t* efw,
+                        WebRtc_Word16* output,
+                        const WebRtc_Word16* nearendClean)
+{
+    int i, j, outCFFT;
+    WebRtc_Word32 tmp32no1;
+
+    // Synthesis
+    for(i = 0, j = 0; i < PART_LEN; i += 4, j += 8)
+    {
+        // We overwrite two more elements in fft[], but it's ok.
+        __asm__("vld2.16 {d20, d21}, [%0, :128]" : : "r"(&(efw[i].real)) : "q10");
+        __asm__("vmov q11, q10" : : : "q10", "q11");
+
+        __asm__("vneg.s16 d23, d23" : : : "q11");
+        __asm__("vst2.16 {d22, d23}, [%0, :128]" : : "r"(&fft[j]): "q11");
+
+        __asm__("vrev64.16 q10, q10" : : : "q10");
+        __asm__("vst2.16 {d20, d21}, [%0]" : : "r"(&fft[PART_LEN4 - j - 6]): "q10");
+    }
+
+    fft[PART_LEN2] = efw[PART_LEN].real;
+    fft[PART_LEN2 + 1] = -efw[PART_LEN].imag;
+
+    // Inverse FFT, result should be scaled with outCFFT.
+    WebRtcSpl_ComplexBitReverse(fft, PART_LEN_SHIFT);
+    outCFFT = WebRtcSpl_ComplexIFFT(fft, PART_LEN_SHIFT, 1);
+
+    // Take only the real values and scale with outCFFT.
+    for (i = 0, j = 0; i < PART_LEN2; i += 8, j+= 16)
+    {
+        __asm__("vld2.16 {d20, d21, d22, d23}, [%0, :256]" : : "r"(&fft[j]) : "q10", "q11");
+        __asm__("vst1.16 {d20, d21}, [%0, :128]" : : "r"(&fft[i]): "q10");
+    }
+
+    int32x4_t tmp32x4_2;
+    __asm__("vdup.32 %q0, %1" : "=w"(tmp32x4_2) : "r"((WebRtc_Word32)
+        (outCFFT - aecm->dfaCleanQDomain)));
+    for (i = 0; i < PART_LEN; i += 4)
+    {
+        int16x4_t tmp16x4_0;
+        int16x4_t tmp16x4_1;
+        int32x4_t tmp32x4_0;
+        int32x4_t tmp32x4_1;
+
+        // fft[i] = (WebRtc_Word16)WEBRTC_SPL_MUL_16_16_RSFT_WITH_ROUND(
+        //        fft[i], WebRtcAecm_kSqrtHanning[i], 14);
+        __asm__("vld1.16 %P0, [%1, :64]" : "=w"(tmp16x4_0) : "r"(&fft[i]));
+        __asm__("vld1.16 %P0, [%1, :64]" : "=w"(tmp16x4_1) : "r"(&WebRtcAecm_kSqrtHanning[i]));
+        __asm__("vmull.s16 %q0, %P1, %P2" : "=w"(tmp32x4_0) : "w"(tmp16x4_0), "w"(tmp16x4_1));
+        __asm__("vrshr.s32 %q0, %q1, #14" : "=w"(tmp32x4_0) : "0"(tmp32x4_0));
+
+        // tmp32no1 = WEBRTC_SPL_SHIFT_W32((WebRtc_Word32)fft[i],
+        //        outCFFT - aecm->dfaCleanQDomain);
+        __asm__("vshl.s32 %q0, %q1, %q2" : "=w"(tmp32x4_0) : "0"(tmp32x4_0), "w"(tmp32x4_2));
+
+        // fft[i] = (WebRtc_Word16)WEBRTC_SPL_SAT(WEBRTC_SPL_WORD16_MAX,
+        //        tmp32no1 + outBuf[i], WEBRTC_SPL_WORD16_MIN);
+        // output[i] = fft[i];
+        __asm__("vld1.16 %P0, [%1, :64]" : "=w"(tmp16x4_0) : "r"(&aecm->outBuf[i]));
+        __asm__("vmovl.s16 %q0, %P1" : "=w"(tmp32x4_1) : "w"(tmp16x4_0));
+        __asm__("vadd.i32 %q0, %q1" : : "w"(tmp32x4_0), "w"(tmp32x4_1));
+        __asm__("vqshrn.s32 %P0, %q1, #0" : "=w"(tmp16x4_0) : "w"(tmp32x4_0));
+        __asm__("vst1.16 %P0, [%1, :64]" : : "w"(tmp16x4_0), "r"(&fft[i]));
+        __asm__("vst1.16 %P0, [%1, :64]" : : "w"(tmp16x4_0), "r"(&output[i]));
+
+        // tmp32no1 = WEBRTC_SPL_MUL_16_16_RSFT(
+        //        fft[PART_LEN + i], WebRtcAecm_kSqrtHanning[PART_LEN - i], 14);
+        __asm__("vld1.16 %P0, [%1, :64]" : "=w"(tmp16x4_0) : "r"(&fft[PART_LEN + i]));
+        __asm__("vld1.16 %P0, [%1, :64]" : "=w"(tmp16x4_1) : "r"(&kSqrtHanningReversed[i]));
+        __asm__("vmull.s16 %q0, %P1, %P2" : "=w"(tmp32x4_0) : "w"(tmp16x4_0), "w"(tmp16x4_1));
+        __asm__("vshr.s32 %q0, %q1, #14" : "=w"(tmp32x4_0) : "0"(tmp32x4_0));
+
+        // tmp32no1 = WEBRTC_SPL_SHIFT_W32(tmp32no1, outCFFT - aecm->dfaCleanQDomain);
+        __asm__("vshl.s32 %q0, %q1, %q2" : "=w"(tmp32x4_0) : "0"(tmp32x4_0), "w"(tmp32x4_2));
+        // outBuf[i] = (WebRtc_Word16)WEBRTC_SPL_SAT(
+        //        WEBRTC_SPL_WORD16_MAX, tmp32no1, WEBRTC_SPL_WORD16_MIN);
+        __asm__("vqshrn.s32 %P0, %q1, #0" : "=w"(tmp16x4_0) : "w"(tmp32x4_0));
+        __asm__("vst1.16 %P0, [%1, :64]" : : "w"(tmp16x4_0), "r"(&aecm->outBuf[i]));
+    }
+
+    // Copy the current block to the old position (outBuf is shifted elsewhere).
+    for (i = 0; i < PART_LEN; i += 16)
+    {
+        __asm__("vld1.16 {d20, d21, d22, d23}, [%0, :256]" : :
+            "r"(&aecm->xBuf[i + PART_LEN]) : "q10");
+        __asm__("vst1.16 {d20, d21, d22, d23}, [%0, :256]" : : "r"(&aecm->xBuf[i]): "q10");
+    }
+    for (i = 0; i < PART_LEN; i += 16)
+    {
+        __asm__("vld1.16 {d20, d21, d22, d23}, [%0, :256]" : :
+            "r"(&aecm->dBufNoisy[i + PART_LEN]) : "q10");
+        __asm__("vst1.16 {d20, d21, d22, d23}, [%0, :256]" : : 
+            "r"(&aecm->dBufNoisy[i]): "q10");
+    }
+    if (nearendClean != NULL) {
+        for (i = 0; i < PART_LEN; i += 16)
+        {
+            __asm__("vld1.16 {d20, d21, d22, d23}, [%0, :256]" : :
+                "r"(&aecm->dBufClean[i + PART_LEN]) : "q10");
+            __asm__("vst1.16 {d20, d21, d22, d23}, [%0, :256]" : :
+                "r"(&aecm->dBufClean[i]): "q10");
+        }
+    }
+}
+
+void WebRtcAecm_CalcLinearEnergies(AecmCore_t* aecm,
+                                   const WebRtc_UWord16* far_spectrum,
+                                   WebRtc_Word32* echo_est,
+                                   WebRtc_UWord32* far_energy,
+                                   WebRtc_UWord32* echo_energy_adapt,
+                                   WebRtc_UWord32* echo_energy_stored)
+{
+    int i;
+
+    register WebRtc_UWord32 far_energy_r;
+    register WebRtc_UWord32 echo_energy_stored_r;
+    register WebRtc_UWord32 echo_energy_adapt_r;
+    uint32x4_t tmp32x4_0;
+
+    __asm__("vmov.i32 q14, #0" : : : "q14"); // far_energy
+    __asm__("vmov.i32 q8,  #0" : : : "q8"); // echo_energy_stored
+    __asm__("vmov.i32 q9,  #0" : : : "q9"); // echo_energy_adapt
+
+    for(i = 0; i < PART_LEN -7; i += 8)
+    {
+        // far_energy += (WebRtc_UWord32)(far_spectrum[i]);
+        __asm__("vld1.16 {d26, d27}, [%0]" : : "r"(&far_spectrum[i]) : "q13");
+        __asm__("vaddw.u16 q14, q14, d26" : : : "q14", "q13");
+        __asm__("vaddw.u16 q14, q14, d27" : : : "q14", "q13");
+
+        // Get estimated echo energies for adaptive channel and stored channel.
+        // echoEst[i] = WEBRTC_SPL_MUL_16_U16(aecm->channelStored[i], far_spectrum[i]);
+        __asm__("vld1.16 {d24, d25}, [%0, :128]" : : "r"(&aecm->channelStored[i]) : "q12");
+        __asm__("vmull.u16 q10, d26, d24" : : : "q12", "q13", "q10");
+        __asm__("vmull.u16 q11, d27, d25" : : : "q12", "q13", "q11");
+        __asm__("vst1.32 {d20, d21, d22, d23}, [%0, :256]" : : "r"(&echo_est[i]):
+            "q10", "q11");
+
+        // echo_energy_stored += (WebRtc_UWord32)echoEst[i];
+        __asm__("vadd.u32 q8, q10" : : : "q10", "q8");
+        __asm__("vadd.u32 q8, q11" : : : "q11", "q8");
+
+        // echo_energy_adapt += WEBRTC_SPL_UMUL_16_16(
+        //     aecm->channelAdapt16[i], far_spectrum[i]);
+        __asm__("vld1.16 {d24, d25}, [%0, :128]" : : "r"(&aecm->channelAdapt16[i]) : "q12");
+        __asm__("vmull.u16 q10, d26, d24" : : : "q12", "q13", "q10");
+        __asm__("vmull.u16 q11, d27, d25" : : : "q12", "q13", "q11");
+        __asm__("vadd.u32 q9, q10" : : : "q9", "q15");
+        __asm__("vadd.u32 q9, q11" : : : "q9", "q11");
+    }
+
+    __asm__("vadd.u32 d28, d29" : : : "q14");
+    __asm__("vpadd.u32 d28, d28" : : : "q14");
+    __asm__("vmov.32 %0, d28[0]" : "=r"(far_energy_r): : "q14");
+
+    __asm__("vadd.u32 d18, d19" : : : "q9");
+    __asm__("vpadd.u32 d18, d18" : : : "q9");
+    __asm__("vmov.32 %0, d18[0]" : "=r"(echo_energy_adapt_r): : "q9");
+
+    __asm__("vadd.u32 d16, d17" : : : "q8");
+    __asm__("vpadd.u32 d16, d16" : : : "q8");
+    __asm__("vmov.32 %0, d16[0]" : "=r"(echo_energy_stored_r): : "q8");
+
+    // Get estimated echo energies for adaptive channel and stored channel.
+    echo_est[i] = WEBRTC_SPL_MUL_16_U16(aecm->channelStored[i], far_spectrum[i]);
+    *echo_energy_stored = echo_energy_stored_r + (WebRtc_UWord32)echo_est[i];
+    *far_energy = far_energy_r + (WebRtc_UWord32)(far_spectrum[i]);
+    *echo_energy_adapt = echo_energy_adapt_r + WEBRTC_SPL_UMUL_16_16(
+        aecm->channelAdapt16[i], far_spectrum[i]);
+}
+
+void WebRtcAecm_StoreAdaptiveChannel(AecmCore_t* aecm,
+                                     const WebRtc_UWord16* far_spectrum,
+                                     WebRtc_Word32* echo_est)
+{
+    int i;
+
+    // During startup we store the channel every block.
+    // Recalculate echo estimate.
+    for(i = 0; i < PART_LEN -7; i += 8)
+    {
+        // aecm->channelStored[i] = acem->channelAdapt16[i];
+        // echo_est[i] = WEBRTC_SPL_MUL_16_U16(aecm->channelStored[i], far_spectrum[i]);
+        __asm__("vld1.16 {d26, d27}, [%0]" : : "r"(&far_spectrum[i]) : "q13");
+        __asm__("vld1.16 {d24, d25}, [%0, :128]" : : "r"(&aecm->channelAdapt16[i]) : "q12");
+        __asm__("vst1.16 {d24, d25}, [%0, :128]" : : "r"(&aecm->channelStored[i]) : "q12");
+        __asm__("vmull.u16 q10, d26, d24" : : : "q12", "q13", "q10");
+        __asm__("vmull.u16 q11, d27, d25" : : : "q12", "q13", "q11");
+        __asm__("vst1.16 {d20, d21, d22, d23}, [%0, :256]" : :
+                               "r"(&echo_est[i]) : "q10", "q11");
+    }
+    aecm->channelStored[i] = aecm->channelAdapt16[i];
+    echo_est[i] = WEBRTC_SPL_MUL_16_U16(aecm->channelStored[i], far_spectrum[i]);
+}
+
+void WebRtcAecm_ResetAdaptiveChannel(AecmCore_t* aecm)
+{
+    int i;
+
+    for(i = 0; i < PART_LEN -7; i += 8)
+    {
+        // aecm->channelAdapt16[i] = aecm->channelStored[i];
+        // aecm->channelAdapt32[i] = WEBRTC_SPL_LSHIFT_W32((WebRtc_Word32)
+        //                           aecm->channelStored[i], 16);
+        __asm__("vld1.16 {d24, d25}, [%0, :128]" : :
+                        "r"(&aecm->channelStored[i]) : "q12");
+        __asm__("vst1.16 {d24, d25}, [%0, :128]" : :
+                        "r"(&aecm->channelAdapt16[i]) : "q12");
+        __asm__("vshll.s16 q10, d24, #16" : : : "q12", "q13", "q10");
+        __asm__("vshll.s16 q11, d25, #16" : : : "q12", "q13", "q11");
+        __asm__("vst1.16 {d20, d21, d22, d23}, [%0, :256]" : :
+                        "r"(&aecm->channelAdapt32[i]): "q10", "q11");
+    }
+    aecm->channelAdapt16[i] = aecm->channelStored[i];
+    aecm->channelAdapt32[i] = WEBRTC_SPL_LSHIFT_W32(
+            (WebRtc_Word32)aecm->channelStored[i], 16);
+}
+
+#endif // #if defined(WEBRTC_ANDROID) && defined(WEBRTC_ARCH_ARM_NEON)