AnotherFoxGuy/webrtc-audio-processing
1
0
Fork 0
Code Pull Requests Activity

Bump to WebRTC M131 release

Browse Source 
Ongoing fixes and improvements, transient suppressor is gone. Also,
dropping isac because it doesn't seem to be useful, and is just build
system deadweight now.

Upstream references:

  Version: 131.0.6778.200
  WebRTC: 79aff54b0fa9238ce3518dd9eaf9610cd6f22e82
  Chromium: 2a19506ad24af755f2a215a4c61f775393e0db42
This commit is contained in:
Arun Raghavan
2024-12-24 19:32:07 -05:00
parent 8bdb53d91c
commit b5c48b97f6
263 changed files with 4628 additions and 20416 deletions
Download Patch File Download Diff File Expand all files Collapse all files

445
webrtc/modules/audio_processing/audio_processing_impl.cc
View File

@ -14,22 +14,23 @@
#include <cstdint>
#include <cstring>
#include <memory>
#include <optional>
#include <string>
#include <type_traits>
#include <utility>
#include "absl/base/nullability.h"
#include "absl/strings/match.h"
#include "absl/strings/string_view.h"
#include "absl/types/optional.h"
#include "api/array_view.h"
#include "api/audio/audio_frame.h"
#include "api/task_queue/task_queue_base.h"
#include "common_audio/audio_converter.h"
#include "common_audio/include/audio_util.h"
#include "modules/audio_processing/aec_dump/aec_dump_factory.h"
#include "modules/audio_processing/audio_buffer.h"
#include "modules/audio_processing/include/audio_frame_view.h"
#include "modules/audio_processing/logging/apm_data_dumper.h"
#include "modules/audio_processing/optionally_built_submodule_creators.h"
#include "rtc_base/checks.h"
#include "rtc_base/experiments/field_trial_parser.h"
#include "rtc_base/logging.h"
@ -322,231 +323,6 @@ constexpr int kUnspecifiedDataDumpInputVolume = -100;
// Throughout webrtc, it's assumed that success is represented by zero.
static_assert(AudioProcessing::kNoError == 0, "kNoError must be zero");
absl::optional<AudioProcessingImpl::GainController2ExperimentParams>
AudioProcessingImpl::GetGainController2ExperimentParams() {
constexpr char kFieldTrialName[] = "WebRTC-Audio-GainController2";
if (!field_trial::IsEnabled(kFieldTrialName)) {
return absl::nullopt;
}
FieldTrialFlag enabled("Enabled", false);
// Whether the gain control should switch to AGC2. Enabled by default.
FieldTrialParameter<bool> switch_to_agc2("switch_to_agc2", true);
// AGC2 input volume controller configuration.
constexpr InputVolumeController::Config kDefaultInputVolumeControllerConfig;
FieldTrialConstrained<int> min_input_volume(
"min_input_volume", kDefaultInputVolumeControllerConfig.min_input_volume,
0, 255);
FieldTrialConstrained<int> clipped_level_min(
"clipped_level_min",
kDefaultInputVolumeControllerConfig.clipped_level_min, 0, 255);
FieldTrialConstrained<int> clipped_level_step(
"clipped_level_step",
kDefaultInputVolumeControllerConfig.clipped_level_step, 0, 255);
FieldTrialConstrained<double> clipped_ratio_threshold(
"clipped_ratio_threshold",
kDefaultInputVolumeControllerConfig.clipped_ratio_threshold, 0, 1);
FieldTrialConstrained<int> clipped_wait_frames(
"clipped_wait_frames",
kDefaultInputVolumeControllerConfig.clipped_wait_frames, 0,
absl::nullopt);
FieldTrialParameter<bool> enable_clipping_predictor(
"enable_clipping_predictor",
kDefaultInputVolumeControllerConfig.enable_clipping_predictor);
FieldTrialConstrained<int> target_range_max_dbfs(
"target_range_max_dbfs",
kDefaultInputVolumeControllerConfig.target_range_max_dbfs, -90, 30);
FieldTrialConstrained<int> target_range_min_dbfs(
"target_range_min_dbfs",
kDefaultInputVolumeControllerConfig.target_range_min_dbfs, -90, 30);
FieldTrialConstrained<int> update_input_volume_wait_frames(
"update_input_volume_wait_frames",
kDefaultInputVolumeControllerConfig.update_input_volume_wait_frames, 0,
absl::nullopt);
FieldTrialConstrained<double> speech_probability_threshold(
"speech_probability_threshold",
kDefaultInputVolumeControllerConfig.speech_probability_threshold, 0, 1);
FieldTrialConstrained<double> speech_ratio_threshold(
"speech_ratio_threshold",
kDefaultInputVolumeControllerConfig.speech_ratio_threshold, 0, 1);
// AGC2 adaptive digital controller configuration.
constexpr AudioProcessing::Config::GainController2::AdaptiveDigital
kDefaultAdaptiveDigitalConfig;
FieldTrialConstrained<double> headroom_db(
"headroom_db", kDefaultAdaptiveDigitalConfig.headroom_db, 0,
absl::nullopt);
FieldTrialConstrained<double> max_gain_db(
"max_gain_db", kDefaultAdaptiveDigitalConfig.max_gain_db, 0,
absl::nullopt);
FieldTrialConstrained<double> initial_gain_db(
"initial_gain_db", kDefaultAdaptiveDigitalConfig.initial_gain_db, 0,
absl::nullopt);
FieldTrialConstrained<double> max_gain_change_db_per_second(
"max_gain_change_db_per_second",
kDefaultAdaptiveDigitalConfig.max_gain_change_db_per_second, 0,
absl::nullopt);
FieldTrialConstrained<double> max_output_noise_level_dbfs(
"max_output_noise_level_dbfs",
kDefaultAdaptiveDigitalConfig.max_output_noise_level_dbfs, absl::nullopt,
0);
// Transient suppressor.
FieldTrialParameter<bool> disallow_transient_suppressor_usage(
"disallow_transient_suppressor_usage", false);
// Field-trial based override for the input volume controller and adaptive
// digital configs.
ParseFieldTrial(
{&enabled, &switch_to_agc2, &min_input_volume, &clipped_level_min,
&clipped_level_step, &clipped_ratio_threshold, &clipped_wait_frames,
&enable_clipping_predictor, &target_range_max_dbfs,
&target_range_min_dbfs, &update_input_volume_wait_frames,
&speech_probability_threshold, &speech_ratio_threshold, &headroom_db,
&max_gain_db, &initial_gain_db, &max_gain_change_db_per_second,
&max_output_noise_level_dbfs, &disallow_transient_suppressor_usage},
field_trial::FindFullName(kFieldTrialName));
// Checked already by `IsEnabled()` before parsing, therefore always true.
RTC_DCHECK(enabled);
const bool do_not_change_agc_config = !switch_to_agc2.Get();
if (do_not_change_agc_config && !disallow_transient_suppressor_usage.Get()) {
// Return an unspecifed value since, in this case, both the AGC2 and TS
// configurations won't be adjusted.
return absl::nullopt;
}
using Params = AudioProcessingImpl::GainController2ExperimentParams;
if (do_not_change_agc_config) {
// Return a value that leaves the AGC2 config unchanged and that always
// disables TS.
return Params{.agc2_config = absl::nullopt,
.disallow_transient_suppressor_usage = true};
}
// Return a value that switches all the gain control to AGC2.
return Params{
.agc2_config =
Params::Agc2Config{
.input_volume_controller =
{
.min_input_volume = min_input_volume.Get(),
.clipped_level_min = clipped_level_min.Get(),
.clipped_level_step = clipped_level_step.Get(),
.clipped_ratio_threshold =
static_cast<float>(clipped_ratio_threshold.Get()),
.clipped_wait_frames = clipped_wait_frames.Get(),
.enable_clipping_predictor =
enable_clipping_predictor.Get(),
.target_range_max_dbfs = target_range_max_dbfs.Get(),
.target_range_min_dbfs = target_range_min_dbfs.Get(),
.update_input_volume_wait_frames =
update_input_volume_wait_frames.Get(),
.speech_probability_threshold = static_cast<float>(
speech_probability_threshold.Get()),
.speech_ratio_threshold =
static_cast<float>(speech_ratio_threshold.Get()),
},
.adaptive_digital_controller =
{
.headroom_db = static_cast<float>(headroom_db.Get()),
.max_gain_db = static_cast<float>(max_gain_db.Get()),
.initial_gain_db =
static_cast<float>(initial_gain_db.Get()),
.max_gain_change_db_per_second = static_cast<float>(
max_gain_change_db_per_second.Get()),
.max_output_noise_level_dbfs =
static_cast<float>(max_output_noise_level_dbfs.Get()),
}},
.disallow_transient_suppressor_usage =
disallow_transient_suppressor_usage.Get()};
}
AudioProcessing::Config AudioProcessingImpl::AdjustConfig(
const AudioProcessing::Config& config,
const absl::optional<AudioProcessingImpl::GainController2ExperimentParams>&
experiment_params) {
if (!experiment_params.has_value() ||
(!experiment_params->agc2_config.has_value() &&
!experiment_params->disallow_transient_suppressor_usage)) {
// When the experiment parameters are unspecified or when the AGC and TS
// configuration are not overridden, return the unmodified configuration.
return config;
}
AudioProcessing::Config adjusted_config = config;
// Override the transient suppressor configuration.
if (experiment_params->disallow_transient_suppressor_usage) {
adjusted_config.transient_suppression.enabled = false;
}
// Override the auto gain control configuration if the AGC1 analog gain
// controller is active and `experiment_params->agc2_config` is specified.
const bool agc1_analog_enabled =
config.gain_controller1.enabled &&
(config.gain_controller1.mode ==
AudioProcessing::Config::GainController1::kAdaptiveAnalog ||
config.gain_controller1.analog_gain_controller.enabled);
if (agc1_analog_enabled && experiment_params->agc2_config.has_value()) {
// Check that the unadjusted AGC config meets the preconditions.
const bool hybrid_agc_config_detected =
config.gain_controller1.enabled &&
config.gain_controller1.analog_gain_controller.enabled &&
!config.gain_controller1.analog_gain_controller
.enable_digital_adaptive &&
config.gain_controller2.enabled &&
config.gain_controller2.adaptive_digital.enabled;
const bool full_agc1_config_detected =
config.gain_controller1.enabled &&
config.gain_controller1.analog_gain_controller.enabled &&
config.gain_controller1.analog_gain_controller
.enable_digital_adaptive &&
!config.gain_controller2.enabled;
const bool one_and_only_one_input_volume_controller =
hybrid_agc_config_detected != full_agc1_config_detected;
const bool agc2_input_volume_controller_enabled =
config.gain_controller2.enabled &&
config.gain_controller2.input_volume_controller.enabled;
if (!one_and_only_one_input_volume_controller ||
agc2_input_volume_controller_enabled) {
RTC_LOG(LS_ERROR) << "Cannot adjust AGC config (precondition failed)";
if (!one_and_only_one_input_volume_controller)
RTC_LOG(LS_ERROR)
<< "One and only one input volume controller must be enabled.";
if (agc2_input_volume_controller_enabled)
RTC_LOG(LS_ERROR)
<< "The AGC2 input volume controller must be disabled.";
} else {
adjusted_config.gain_controller1.enabled = false;
adjusted_config.gain_controller1.analog_gain_controller.enabled = false;
adjusted_config.gain_controller2.enabled = true;
adjusted_config.gain_controller2.input_volume_controller.enabled = true;
adjusted_config.gain_controller2.adaptive_digital =
experiment_params->agc2_config->adaptive_digital_controller;
adjusted_config.gain_controller2.adaptive_digital.enabled = true;
}
}
return adjusted_config;
}
bool AudioProcessingImpl::UseApmVadSubModule(
const AudioProcessing::Config& config,
const absl::optional<GainController2ExperimentParams>& experiment_params) {
// The VAD as an APM sub-module is needed only in one case, that is when TS
// and AGC2 are both enabled and when the AGC2 experiment is running and its
// parameters require to fully switch the gain control to AGC2.
return config.transient_suppression.enabled &&
config.gain_controller2.enabled &&
(config.gain_controller2.input_volume_controller.enabled ||
config.gain_controller2.adaptive_digital.enabled) &&
experiment_params.has_value() &&
experiment_params->agc2_config.has_value();
}
AudioProcessingImpl::SubmoduleStates::SubmoduleStates(
bool capture_post_processor_enabled,
bool render_pre_processor_enabled,
@ -561,10 +337,8 @@ bool AudioProcessingImpl::SubmoduleStates::Update(
bool noise_suppressor_enabled,
bool adaptive_gain_controller_enabled,
bool gain_controller2_enabled,
bool voice_activity_detector_enabled,
bool gain_adjustment_enabled,
bool echo_controller_enabled,
bool transient_suppressor_enabled) {
bool echo_controller_enabled) {
bool changed = false;
changed |= (high_pass_filter_enabled != high_pass_filter_enabled_);
changed |=
@ -573,21 +347,16 @@ bool AudioProcessingImpl::SubmoduleStates::Update(
changed |=
(adaptive_gain_controller_enabled != adaptive_gain_controller_enabled_);
changed |= (gain_controller2_enabled != gain_controller2_enabled_);
changed |=
(voice_activity_detector_enabled != voice_activity_detector_enabled_);
changed |= (gain_adjustment_enabled != gain_adjustment_enabled_);
changed |= (echo_controller_enabled != echo_controller_enabled_);
changed |= (transient_suppressor_enabled != transient_suppressor_enabled_);
if (changed) {
high_pass_filter_enabled_ = high_pass_filter_enabled;
mobile_echo_controller_enabled_ = mobile_echo_controller_enabled;
noise_suppressor_enabled_ = noise_suppressor_enabled;
adaptive_gain_controller_enabled_ = adaptive_gain_controller_enabled;
gain_controller2_enabled_ = gain_controller2_enabled;
voice_activity_detector_enabled_ = voice_activity_detector_enabled;
gain_adjustment_enabled_ = gain_adjustment_enabled;
echo_controller_enabled_ = echo_controller_enabled;
transient_suppressor_enabled_ = transient_suppressor_enabled;
}
changed |= first_update_;
@ -664,14 +433,12 @@ AudioProcessingImpl::AudioProcessingImpl(
: data_dumper_(new ApmDataDumper(instance_count_.fetch_add(1) + 1)),
use_setup_specific_default_aec3_config_(
UseSetupSpecificDefaultAec3Congfig()),
gain_controller2_experiment_params_(GetGainController2ExperimentParams()),
transient_suppressor_vad_mode_(TransientSuppressor::VadMode::kDefault),
capture_runtime_settings_(RuntimeSettingQueueSize()),
render_runtime_settings_(RuntimeSettingQueueSize()),
capture_runtime_settings_enqueuer_(&capture_runtime_settings_),
render_runtime_settings_enqueuer_(&render_runtime_settings_),
echo_control_factory_(std::move(echo_control_factory)),
config_(AdjustConfig(config, gain_controller2_experiment_params_)),
config_(config),
submodule_states_(!!capture_post_processor,
!!render_pre_processor,
!!capture_analyzer),
@ -684,8 +451,7 @@ AudioProcessingImpl::AudioProcessingImpl(
!field_trial::IsEnabled(
"WebRTC-ApmExperimentalMultiChannelCaptureKillSwitch"),
EnforceSplitBandHpf(),
MinimizeProcessingForUnusedOutput(),
field_trial::IsEnabled("WebRTC-TransientSuppressorForcedOff")),
MinimizeProcessingForUnusedOutput()),
capture_(),
capture_nonlocked_(),
applied_input_volume_stats_reporter_(
@ -806,12 +572,10 @@ void AudioProcessingImpl::InitializeLocked() {
AllocateRenderQueue();
InitializeGainController1();
InitializeTransientSuppressor();
InitializeHighPassFilter(true);
InitializeResidualEchoDetector();
InitializeEchoController();
InitializeGainController2();
InitializeVoiceActivityDetector();
InitializeNoiseSuppressor();
InitializeAnalyzer();
InitializePostProcessor();
@ -906,52 +670,41 @@ void AudioProcessingImpl::ApplyConfig(const AudioProcessing::Config& config) {
MutexLock lock_render(&mutex_render_);
MutexLock lock_capture(&mutex_capture_);
const auto adjusted_config =
AdjustConfig(config, gain_controller2_experiment_params_);
RTC_LOG(LS_INFO) << "AudioProcessing::ApplyConfig: "
<< adjusted_config.ToString();
RTC_LOG(LS_INFO) << "AudioProcessing::ApplyConfig: " << config.ToString();
const bool pipeline_config_changed =
config_.pipeline.multi_channel_render !=
adjusted_config.pipeline.multi_channel_render ||
config.pipeline.multi_channel_render ||
config_.pipeline.multi_channel_capture !=
adjusted_config.pipeline.multi_channel_capture ||
config.pipeline.multi_channel_capture ||
config_.pipeline.maximum_internal_processing_rate !=
adjusted_config.pipeline.maximum_internal_processing_rate ||
config.pipeline.maximum_internal_processing_rate ||
config_.pipeline.capture_downmix_method !=
adjusted_config.pipeline.capture_downmix_method;
config.pipeline.capture_downmix_method;
const bool aec_config_changed =
config_.echo_canceller.enabled !=
adjusted_config.echo_canceller.enabled ||
config_.echo_canceller.mobile_mode !=
adjusted_config.echo_canceller.mobile_mode;
config_.echo_canceller.enabled != config.echo_canceller.enabled ||
config_.echo_canceller.mobile_mode != config.echo_canceller.mobile_mode;
const bool agc1_config_changed =
config_.gain_controller1 != adjusted_config.gain_controller1;
config_.gain_controller1 != config.gain_controller1;
const bool agc2_config_changed =
config_.gain_controller2 != adjusted_config.gain_controller2;
config_.gain_controller2 != config.gain_controller2;
const bool ns_config_changed =
config_.noise_suppression.enabled !=
adjusted_config.noise_suppression.enabled ||
config_.noise_suppression.level !=
adjusted_config.noise_suppression.level;
const bool ts_config_changed = config_.transient_suppression.enabled !=
adjusted_config.transient_suppression.enabled;
config_.noise_suppression.enabled != config.noise_suppression.enabled ||
config_.noise_suppression.level != config.noise_suppression.level;
const bool pre_amplifier_config_changed =
config_.pre_amplifier.enabled != adjusted_config.pre_amplifier.enabled ||
config_.pre_amplifier.enabled != config.pre_amplifier.enabled ||
config_.pre_amplifier.fixed_gain_factor !=
adjusted_config.pre_amplifier.fixed_gain_factor;
config.pre_amplifier.fixed_gain_factor;
const bool gain_adjustment_config_changed =
config_.capture_level_adjustment !=
adjusted_config.capture_level_adjustment;
config_.capture_level_adjustment != config.capture_level_adjustment;
config_ = adjusted_config;
config_ = config;
if (aec_config_changed) {
InitializeEchoController();
@ -961,10 +714,6 @@ void AudioProcessingImpl::ApplyConfig(const AudioProcessing::Config& config) {
InitializeNoiseSuppressor();
}
if (ts_config_changed) {
InitializeTransientSuppressor();
}
InitializeHighPassFilter(false);
if (agc1_config_changed) {
@ -978,11 +727,8 @@ void AudioProcessingImpl::ApplyConfig(const AudioProcessing::Config& config) {
config_.gain_controller2 = AudioProcessing::Config::GainController2();
}
if (agc2_config_changed || ts_config_changed) {
// AGC2 also depends on TS because of the possible dependency on the APM VAD
// sub-module.
if (agc2_config_changed) {
InitializeGainController2();
InitializeVoiceActivityDetector();
}
if (pre_amplifier_config_changed || gain_adjustment_config_changed) {
@ -996,12 +742,6 @@ void AudioProcessingImpl::ApplyConfig(const AudioProcessing::Config& config) {
}
}
void AudioProcessingImpl::OverrideSubmoduleCreationForTesting(
const ApmSubmoduleCreationOverrides& overrides) {
MutexLock lock(&mutex_capture_);
submodule_creation_overrides_ = overrides;
}
int AudioProcessingImpl::proc_sample_rate_hz() const {
// Used as callback from submodules, hence locking is not allowed.
return capture_nonlocked_.capture_processing_format.sample_rate_hz();
@ -1660,7 +1400,7 @@ int AudioProcessingImpl::ProcessCaptureStreamLocked() {
if (submodules_.agc_manager) {
submodules_.agc_manager->Process(*capture_buffer);
absl::optional<int> new_digital_gain =
std::optional<int> new_digital_gain =
submodules_.agc_manager->GetDigitalComressionGain();
if (new_digital_gain && submodules_.gain_control) {
submodules_.gain_control->set_compression_gain_db(*new_digital_gain);
@ -1697,44 +1437,6 @@ int AudioProcessingImpl::ProcessCaptureStreamLocked() {
capture_buffer->num_frames()));
}
absl::optional<float> voice_probability;
if (!!submodules_.voice_activity_detector) {
voice_probability = submodules_.voice_activity_detector->Analyze(
AudioFrameView<const float>(capture_buffer->channels(),
capture_buffer->num_channels(),
capture_buffer->num_frames()));
}
if (submodules_.transient_suppressor) {
float transient_suppressor_voice_probability = 1.0f;
switch (transient_suppressor_vad_mode_) {
case TransientSuppressor::VadMode::kDefault:
if (submodules_.agc_manager) {
transient_suppressor_voice_probability =
submodules_.agc_manager->voice_probability();
}
break;
case TransientSuppressor::VadMode::kRnnVad:
RTC_DCHECK(voice_probability.has_value());
transient_suppressor_voice_probability = *voice_probability;
break;
case TransientSuppressor::VadMode::kNoVad:
// The transient suppressor will ignore `voice_probability`.
break;
}
float delayed_voice_probability =
submodules_.transient_suppressor->Suppress(
capture_buffer->channels()[0], capture_buffer->num_frames(),
capture_buffer->num_channels(),
capture_buffer->split_bands_const(0)[kBand0To8kHz],
capture_buffer->num_frames_per_band(),
/*reference_data=*/nullptr, /*reference_length=*/0,
transient_suppressor_voice_probability, capture_.key_pressed);
if (voice_probability.has_value()) {
*voice_probability = delayed_voice_probability;
}
}
// Experimental APM sub-module that analyzes `capture_buffer`.
if (submodules_.capture_analyzer) {
submodules_.capture_analyzer->Analyze(capture_buffer);
@ -1744,8 +1446,8 @@ int AudioProcessingImpl::ProcessCaptureStreamLocked() {
// TODO(bugs.webrtc.org/7494): Let AGC2 detect applied input volume
// changes.
submodules_.gain_controller2->Process(
voice_probability, capture_.applied_input_volume_changed,
capture_buffer);
/*speech_probability=*/std::nullopt,
capture_.applied_input_volume_changed, capture_buffer);
}
if (submodules_.capture_post_processor) {
@ -2023,7 +1725,7 @@ void AudioProcessingImpl::set_stream_analog_level_locked(int level) {
// Invalidate any previously recommended input volume which will be updated by
// `ProcessStream()`.
capture_.recommended_input_volume = absl::nullopt;
capture_.recommended_input_volume = std::nullopt;
if (submodules_.agc_manager) {
submodules_.agc_manager->set_stream_analog_level(level);
@ -2056,7 +1758,7 @@ void AudioProcessingImpl::UpdateRecommendedInputVolumeLocked() {
if (!capture_.applied_input_volume.has_value()) {
// When `set_stream_analog_level()` is not called, no input level can be
// recommended.
capture_.recommended_input_volume = absl::nullopt;
capture_.recommended_input_volume = std::nullopt;
return;
}
@ -2082,9 +1784,10 @@ void AudioProcessingImpl::UpdateRecommendedInputVolumeLocked() {
capture_.recommended_input_volume = capture_.applied_input_volume;
}
bool AudioProcessingImpl::CreateAndAttachAecDump(absl::string_view file_name,
int64_t max_log_size_bytes,
rtc::TaskQueue* worker_queue) {
bool AudioProcessingImpl::CreateAndAttachAecDump(
absl::string_view file_name,
int64_t max_log_size_bytes,
absl::Nonnull<TaskQueueBase*> worker_queue) {
std::unique_ptr<AecDump> aec_dump =
AecDumpFactory::Create(file_name, max_log_size_bytes, worker_queue);
if (!aec_dump) {
@ -2095,9 +1798,10 @@ bool AudioProcessingImpl::CreateAndAttachAecDump(absl::string_view file_name,
return true;
}
bool AudioProcessingImpl::CreateAndAttachAecDump(FILE* handle,
int64_t max_log_size_bytes,
rtc::TaskQueue* worker_queue) {
bool AudioProcessingImpl::CreateAndAttachAecDump(
FILE* handle,
int64_t max_log_size_bytes,
absl::Nonnull<TaskQueueBase*> worker_queue) {
std::unique_ptr<AecDump> aec_dump =
AecDumpFactory::Create(handle, max_log_size_bytes, worker_queue);
if (!aec_dump) {
@ -2142,43 +1846,9 @@ bool AudioProcessingImpl::UpdateActiveSubmoduleStates() {
return submodule_states_.Update(
config_.high_pass_filter.enabled, !!submodules_.echo_control_mobile,
!!submodules_.noise_suppressor, !!submodules_.gain_control,
!!submodules_.gain_controller2, !!submodules_.voice_activity_detector,
!!submodules_.gain_controller2,
config_.pre_amplifier.enabled || config_.capture_level_adjustment.enabled,
capture_nonlocked_.echo_controller_enabled,
!!submodules_.transient_suppressor);
}
void AudioProcessingImpl::InitializeTransientSuppressor() {
// Choose the VAD mode for TS and detect a VAD mode change.
const TransientSuppressor::VadMode previous_vad_mode =
transient_suppressor_vad_mode_;
transient_suppressor_vad_mode_ = TransientSuppressor::VadMode::kDefault;
if (UseApmVadSubModule(config_, gain_controller2_experiment_params_)) {
transient_suppressor_vad_mode_ = TransientSuppressor::VadMode::kRnnVad;
}
const bool vad_mode_changed =
previous_vad_mode != transient_suppressor_vad_mode_;
if (config_.transient_suppression.enabled &&
!constants_.transient_suppressor_forced_off) {
// Attempt to create a transient suppressor, if one is not already created.
if (!submodules_.transient_suppressor || vad_mode_changed) {
submodules_.transient_suppressor = CreateTransientSuppressor(
submodule_creation_overrides_, transient_suppressor_vad_mode_,
proc_fullband_sample_rate_hz(), capture_nonlocked_.split_rate,
num_proc_channels());
if (!submodules_.transient_suppressor) {
RTC_LOG(LS_WARNING)
<< "No transient suppressor created (probably disabled)";
}
} else {
submodules_.transient_suppressor->Initialize(
proc_fullband_sample_rate_hz(), capture_nonlocked_.split_rate,
num_proc_channels());
}
} else {
submodules_.transient_suppressor.reset();
}
capture_nonlocked_.echo_controller_enabled);
}
void AudioProcessingImpl::InitializeHighPassFilter(bool forced_reset) {
@ -2220,7 +1890,7 @@ void AudioProcessingImpl::InitializeEchoController() {
RTC_DCHECK(submodules_.echo_controller);
} else {
EchoCanceller3Config config;
absl::optional<EchoCanceller3Config> multichannel_config;
std::optional<EchoCanceller3Config> multichannel_config;
if (use_setup_specific_default_aec3_config_) {
multichannel_config = EchoCanceller3::CreateDefaultMultichannelConfig();
}
@ -2361,46 +2031,19 @@ void AudioProcessingImpl::InitializeGainController2() {
submodules_.gain_controller2.reset();
return;
}
// Override the input volume controller configuration if the AGC2 experiment
// is running and its parameters require to fully switch the gain control to
// Input volume controller configuration if the AGC2 is running
// and its parameters require to fully switch the gain control to
// AGC2.
const bool input_volume_controller_config_overridden =
gain_controller2_experiment_params_.has_value() &&
gain_controller2_experiment_params_->agc2_config.has_value();
const InputVolumeController::Config input_volume_controller_config =
input_volume_controller_config_overridden
? gain_controller2_experiment_params_->agc2_config
->input_volume_controller
: InputVolumeController::Config{};
// If the APM VAD sub-module is not used, let AGC2 use its internal VAD.
const bool use_internal_vad =
!UseApmVadSubModule(config_, gain_controller2_experiment_params_);
InputVolumeController::Config{};
submodules_.gain_controller2 = std::make_unique<GainController2>(
config_.gain_controller2, input_volume_controller_config,
proc_fullband_sample_rate_hz(), num_proc_channels(), use_internal_vad);
proc_fullband_sample_rate_hz(), num_output_channels(),
/*use_internal_vad=*/true);
submodules_.gain_controller2->SetCaptureOutputUsed(
capture_.capture_output_used);
}
void AudioProcessingImpl::InitializeVoiceActivityDetector() {
if (!UseApmVadSubModule(config_, gain_controller2_experiment_params_)) {
submodules_.voice_activity_detector.reset();
return;
}
if (!submodules_.voice_activity_detector) {
RTC_DCHECK(!!submodules_.gain_controller2);
// TODO(bugs.webrtc.org/13663): Cache CPU features in APM and use here.
submodules_.voice_activity_detector =
std::make_unique<VoiceActivityDetectorWrapper>(
submodules_.gain_controller2->GetCpuFeatures(),
proc_fullband_sample_rate_hz());
} else {
submodules_.voice_activity_detector->Initialize(
proc_fullband_sample_rate_hz());
}
}
void AudioProcessingImpl::InitializeNoiseSuppressor() {
submodules_.noise_suppressor.reset();
@ -2535,8 +2178,6 @@ void AudioProcessingImpl::WriteAecDumpConfigMessage(bool forced) {
apm_config.ns_enabled = config_.noise_suppression.enabled;
apm_config.ns_level = static_cast<int>(config_.noise_suppression.level);
apm_config.transient_suppression_enabled =
config_.transient_suppression.enabled;
apm_config.experiments_description = experiments_description;
apm_config.pre_amplifier_enabled = config_.pre_amplifier.enabled;
apm_config.pre_amplifier_fixed_gain_factor =