mirror of
https://github.com/mollyim/webrtc.git
synced 2025-05-15 23:01:21 +01:00

After the refined filter has been determined to perform better than the coarse filter, and the coefficients of the coarse filters are overwritten by the ones from the refined filter, at least 100 ms have to pass before the adaptation of the refined filter is allowed to speed up due to good coarse filter performance. This change solves the vicious circle described in webrtc:12265, where the coarse and refined filters can diverge over time. This feature can be disabled remotely via a kill-switch. When disabled the AEC output is bit-exact to before the change. Bug: webrtc:12265,chromium:1155477 Change-Id: Iacd6e325e987dd8a475bb3e8163fee714c65b20a Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/196501 Reviewed-by: Per Åhgren <peah@webrtc.org> Commit-Queue: Gustaf Ullberg <gustaf@webrtc.org> Cr-Commit-Position: refs/heads/master@{#32801}
486 lines
19 KiB
C++
486 lines
19 KiB
C++
/*
|
|
* Copyright (c) 2017 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 "modules/audio_processing/aec3/aec_state.h"
|
|
|
|
#include <math.h>
|
|
|
|
#include <algorithm>
|
|
#include <numeric>
|
|
#include <vector>
|
|
|
|
#include "absl/types/optional.h"
|
|
#include "api/array_view.h"
|
|
#include "modules/audio_processing/aec3/aec3_common.h"
|
|
#include "modules/audio_processing/logging/apm_data_dumper.h"
|
|
#include "rtc_base/atomic_ops.h"
|
|
#include "rtc_base/checks.h"
|
|
#include "system_wrappers/include/field_trial.h"
|
|
|
|
namespace webrtc {
|
|
namespace {
|
|
|
|
bool DeactivateInitialStateResetAtEchoPathChange() {
|
|
return field_trial::IsEnabled(
|
|
"WebRTC-Aec3DeactivateInitialStateResetKillSwitch");
|
|
}
|
|
|
|
bool FullResetAtEchoPathChange() {
|
|
return !field_trial::IsEnabled("WebRTC-Aec3AecStateFullResetKillSwitch");
|
|
}
|
|
|
|
bool SubtractorAnalyzerResetAtEchoPathChange() {
|
|
return !field_trial::IsEnabled(
|
|
"WebRTC-Aec3AecStateSubtractorAnalyzerResetKillSwitch");
|
|
}
|
|
|
|
void ComputeAvgRenderReverb(
|
|
const SpectrumBuffer& spectrum_buffer,
|
|
int delay_blocks,
|
|
float reverb_decay,
|
|
ReverbModel* reverb_model,
|
|
rtc::ArrayView<float, kFftLengthBy2Plus1> reverb_power_spectrum) {
|
|
RTC_DCHECK(reverb_model);
|
|
const size_t num_render_channels = spectrum_buffer.buffer[0].size();
|
|
int idx_at_delay =
|
|
spectrum_buffer.OffsetIndex(spectrum_buffer.read, delay_blocks);
|
|
int idx_past = spectrum_buffer.IncIndex(idx_at_delay);
|
|
|
|
std::array<float, kFftLengthBy2Plus1> X2_data;
|
|
rtc::ArrayView<const float> X2;
|
|
if (num_render_channels > 1) {
|
|
auto average_channels =
|
|
[](size_t num_render_channels,
|
|
rtc::ArrayView<const std::array<float, kFftLengthBy2Plus1>>
|
|
spectrum_band_0,
|
|
rtc::ArrayView<float, kFftLengthBy2Plus1> render_power) {
|
|
std::fill(render_power.begin(), render_power.end(), 0.f);
|
|
for (size_t ch = 0; ch < num_render_channels; ++ch) {
|
|
for (size_t k = 0; k < kFftLengthBy2Plus1; ++k) {
|
|
render_power[k] += spectrum_band_0[ch][k];
|
|
}
|
|
}
|
|
const float normalizer = 1.f / num_render_channels;
|
|
for (size_t k = 0; k < kFftLengthBy2Plus1; ++k) {
|
|
render_power[k] *= normalizer;
|
|
}
|
|
};
|
|
average_channels(num_render_channels, spectrum_buffer.buffer[idx_past],
|
|
X2_data);
|
|
reverb_model->UpdateReverbNoFreqShaping(
|
|
X2_data, /*power_spectrum_scaling=*/1.0f, reverb_decay);
|
|
|
|
average_channels(num_render_channels, spectrum_buffer.buffer[idx_at_delay],
|
|
X2_data);
|
|
X2 = X2_data;
|
|
} else {
|
|
reverb_model->UpdateReverbNoFreqShaping(
|
|
spectrum_buffer.buffer[idx_past][/*channel=*/0],
|
|
/*power_spectrum_scaling=*/1.0f, reverb_decay);
|
|
|
|
X2 = spectrum_buffer.buffer[idx_at_delay][/*channel=*/0];
|
|
}
|
|
|
|
rtc::ArrayView<const float, kFftLengthBy2Plus1> reverb_power =
|
|
reverb_model->reverb();
|
|
for (size_t k = 0; k < X2.size(); ++k) {
|
|
reverb_power_spectrum[k] = X2[k] + reverb_power[k];
|
|
}
|
|
}
|
|
|
|
} // namespace
|
|
|
|
int AecState::instance_count_ = 0;
|
|
|
|
void AecState::GetResidualEchoScaling(
|
|
rtc::ArrayView<float> residual_scaling) const {
|
|
bool filter_has_had_time_to_converge;
|
|
if (config_.filter.conservative_initial_phase) {
|
|
filter_has_had_time_to_converge =
|
|
strong_not_saturated_render_blocks_ >= 1.5f * kNumBlocksPerSecond;
|
|
} else {
|
|
filter_has_had_time_to_converge =
|
|
strong_not_saturated_render_blocks_ >= 0.8f * kNumBlocksPerSecond;
|
|
}
|
|
echo_audibility_.GetResidualEchoScaling(filter_has_had_time_to_converge,
|
|
residual_scaling);
|
|
}
|
|
|
|
absl::optional<float> AecState::ErleUncertainty() const {
|
|
if (SaturatedEcho()) {
|
|
return 1.f;
|
|
}
|
|
|
|
return absl::nullopt;
|
|
}
|
|
|
|
AecState::AecState(const EchoCanceller3Config& config,
|
|
size_t num_capture_channels)
|
|
: data_dumper_(
|
|
new ApmDataDumper(rtc::AtomicOps::Increment(&instance_count_))),
|
|
config_(config),
|
|
num_capture_channels_(num_capture_channels),
|
|
deactivate_initial_state_reset_at_echo_path_change_(
|
|
DeactivateInitialStateResetAtEchoPathChange()),
|
|
full_reset_at_echo_path_change_(FullResetAtEchoPathChange()),
|
|
subtractor_analyzer_reset_at_echo_path_change_(
|
|
SubtractorAnalyzerResetAtEchoPathChange()),
|
|
initial_state_(config_),
|
|
delay_state_(config_, num_capture_channels_),
|
|
transparent_state_(TransparentMode::Create(config_)),
|
|
filter_quality_state_(config_, num_capture_channels_),
|
|
erl_estimator_(2 * kNumBlocksPerSecond),
|
|
erle_estimator_(2 * kNumBlocksPerSecond, config_, num_capture_channels_),
|
|
filter_analyzer_(config_, num_capture_channels_),
|
|
echo_audibility_(
|
|
config_.echo_audibility.use_stationarity_properties_at_init),
|
|
reverb_model_estimator_(config_, num_capture_channels_),
|
|
subtractor_output_analyzer_(num_capture_channels_) {}
|
|
|
|
AecState::~AecState() = default;
|
|
|
|
void AecState::HandleEchoPathChange(
|
|
const EchoPathVariability& echo_path_variability) {
|
|
const auto full_reset = [&]() {
|
|
filter_analyzer_.Reset();
|
|
capture_signal_saturation_ = false;
|
|
strong_not_saturated_render_blocks_ = 0;
|
|
blocks_with_active_render_ = 0;
|
|
if (!deactivate_initial_state_reset_at_echo_path_change_) {
|
|
initial_state_.Reset();
|
|
}
|
|
if (transparent_state_) {
|
|
transparent_state_->Reset();
|
|
}
|
|
erle_estimator_.Reset(true);
|
|
erl_estimator_.Reset();
|
|
filter_quality_state_.Reset();
|
|
};
|
|
|
|
// TODO(peah): Refine the reset scheme according to the type of gain and
|
|
// delay adjustment.
|
|
|
|
if (full_reset_at_echo_path_change_ &&
|
|
echo_path_variability.delay_change !=
|
|
EchoPathVariability::DelayAdjustment::kNone) {
|
|
full_reset();
|
|
} else if (echo_path_variability.gain_change) {
|
|
erle_estimator_.Reset(false);
|
|
}
|
|
if (subtractor_analyzer_reset_at_echo_path_change_) {
|
|
subtractor_output_analyzer_.HandleEchoPathChange();
|
|
}
|
|
}
|
|
|
|
void AecState::Update(
|
|
const absl::optional<DelayEstimate>& external_delay,
|
|
rtc::ArrayView<const std::vector<std::array<float, kFftLengthBy2Plus1>>>
|
|
adaptive_filter_frequency_responses,
|
|
rtc::ArrayView<const std::vector<float>> adaptive_filter_impulse_responses,
|
|
const RenderBuffer& render_buffer,
|
|
rtc::ArrayView<const std::array<float, kFftLengthBy2Plus1>> E2_refined,
|
|
rtc::ArrayView<const std::array<float, kFftLengthBy2Plus1>> Y2,
|
|
rtc::ArrayView<const SubtractorOutput> subtractor_output) {
|
|
RTC_DCHECK_EQ(num_capture_channels_, Y2.size());
|
|
RTC_DCHECK_EQ(num_capture_channels_, subtractor_output.size());
|
|
RTC_DCHECK_EQ(num_capture_channels_,
|
|
adaptive_filter_frequency_responses.size());
|
|
RTC_DCHECK_EQ(num_capture_channels_,
|
|
adaptive_filter_impulse_responses.size());
|
|
|
|
// Analyze the filter outputs and filters.
|
|
bool any_filter_converged;
|
|
bool any_coarse_filter_converged;
|
|
bool all_filters_diverged;
|
|
subtractor_output_analyzer_.Update(subtractor_output, &any_filter_converged,
|
|
&any_coarse_filter_converged,
|
|
&all_filters_diverged);
|
|
|
|
bool any_filter_consistent;
|
|
float max_echo_path_gain;
|
|
filter_analyzer_.Update(adaptive_filter_impulse_responses, render_buffer,
|
|
&any_filter_consistent, &max_echo_path_gain);
|
|
|
|
// Estimate the direct path delay of the filter.
|
|
if (config_.filter.use_linear_filter) {
|
|
delay_state_.Update(filter_analyzer_.FilterDelaysBlocks(), external_delay,
|
|
strong_not_saturated_render_blocks_);
|
|
}
|
|
|
|
const std::vector<std::vector<float>>& aligned_render_block =
|
|
render_buffer.Block(-delay_state_.MinDirectPathFilterDelay())[0];
|
|
|
|
// Update render counters.
|
|
bool active_render = false;
|
|
for (size_t ch = 0; ch < aligned_render_block.size(); ++ch) {
|
|
const float render_energy = std::inner_product(
|
|
aligned_render_block[ch].begin(), aligned_render_block[ch].end(),
|
|
aligned_render_block[ch].begin(), 0.f);
|
|
if (render_energy > (config_.render_levels.active_render_limit *
|
|
config_.render_levels.active_render_limit) *
|
|
kFftLengthBy2) {
|
|
active_render = true;
|
|
break;
|
|
}
|
|
}
|
|
blocks_with_active_render_ += active_render ? 1 : 0;
|
|
strong_not_saturated_render_blocks_ +=
|
|
active_render && !SaturatedCapture() ? 1 : 0;
|
|
|
|
std::array<float, kFftLengthBy2Plus1> avg_render_spectrum_with_reverb;
|
|
|
|
ComputeAvgRenderReverb(render_buffer.GetSpectrumBuffer(),
|
|
delay_state_.MinDirectPathFilterDelay(), ReverbDecay(),
|
|
&avg_render_reverb_, avg_render_spectrum_with_reverb);
|
|
|
|
if (config_.echo_audibility.use_stationarity_properties) {
|
|
// Update the echo audibility evaluator.
|
|
echo_audibility_.Update(render_buffer, avg_render_reverb_.reverb(),
|
|
delay_state_.MinDirectPathFilterDelay(),
|
|
delay_state_.ExternalDelayReported());
|
|
}
|
|
|
|
// Update the ERL and ERLE measures.
|
|
if (initial_state_.TransitionTriggered()) {
|
|
erle_estimator_.Reset(false);
|
|
}
|
|
|
|
erle_estimator_.Update(render_buffer, adaptive_filter_frequency_responses,
|
|
avg_render_spectrum_with_reverb, Y2, E2_refined,
|
|
subtractor_output_analyzer_.ConvergedFilters());
|
|
|
|
erl_estimator_.Update(
|
|
subtractor_output_analyzer_.ConvergedFilters(),
|
|
render_buffer.Spectrum(delay_state_.MinDirectPathFilterDelay()), Y2);
|
|
|
|
// Detect and flag echo saturation.
|
|
if (config_.ep_strength.echo_can_saturate) {
|
|
saturation_detector_.Update(aligned_render_block, SaturatedCapture(),
|
|
UsableLinearEstimate(), subtractor_output,
|
|
max_echo_path_gain);
|
|
} else {
|
|
RTC_DCHECK(!saturation_detector_.SaturatedEcho());
|
|
}
|
|
|
|
// Update the decision on whether to use the initial state parameter set.
|
|
initial_state_.Update(active_render, SaturatedCapture());
|
|
|
|
// Detect whether the transparent mode should be activated.
|
|
if (transparent_state_) {
|
|
transparent_state_->Update(
|
|
delay_state_.MinDirectPathFilterDelay(), any_filter_consistent,
|
|
any_filter_converged, any_coarse_filter_converged, all_filters_diverged,
|
|
active_render, SaturatedCapture());
|
|
}
|
|
|
|
// Analyze the quality of the filter.
|
|
filter_quality_state_.Update(active_render, TransparentModeActive(),
|
|
SaturatedCapture(), external_delay,
|
|
any_filter_converged);
|
|
|
|
// Update the reverb estimate.
|
|
const bool stationary_block =
|
|
config_.echo_audibility.use_stationarity_properties &&
|
|
echo_audibility_.IsBlockStationary();
|
|
|
|
reverb_model_estimator_.Update(
|
|
filter_analyzer_.GetAdjustedFilters(),
|
|
adaptive_filter_frequency_responses,
|
|
erle_estimator_.GetInstLinearQualityEstimates(),
|
|
delay_state_.DirectPathFilterDelays(),
|
|
filter_quality_state_.UsableLinearFilterOutputs(), stationary_block);
|
|
|
|
erle_estimator_.Dump(data_dumper_);
|
|
reverb_model_estimator_.Dump(data_dumper_.get());
|
|
data_dumper_->DumpRaw("aec3_active_render", active_render);
|
|
data_dumper_->DumpRaw("aec3_erl", Erl());
|
|
data_dumper_->DumpRaw("aec3_erl_time_domain", ErlTimeDomain());
|
|
data_dumper_->DumpRaw("aec3_erle", Erle()[0]);
|
|
data_dumper_->DumpRaw("aec3_usable_linear_estimate", UsableLinearEstimate());
|
|
data_dumper_->DumpRaw("aec3_transparent_mode", TransparentModeActive());
|
|
data_dumper_->DumpRaw("aec3_filter_delay",
|
|
filter_analyzer_.MinFilterDelayBlocks());
|
|
|
|
data_dumper_->DumpRaw("aec3_any_filter_consistent", any_filter_consistent);
|
|
data_dumper_->DumpRaw("aec3_initial_state",
|
|
initial_state_.InitialStateActive());
|
|
data_dumper_->DumpRaw("aec3_capture_saturation", SaturatedCapture());
|
|
data_dumper_->DumpRaw("aec3_echo_saturation", SaturatedEcho());
|
|
data_dumper_->DumpRaw("aec3_any_filter_converged", any_filter_converged);
|
|
data_dumper_->DumpRaw("aec3_any_coarse_filter_converged",
|
|
any_coarse_filter_converged);
|
|
data_dumper_->DumpRaw("aec3_all_filters_diverged", all_filters_diverged);
|
|
|
|
data_dumper_->DumpRaw("aec3_external_delay_avaliable",
|
|
external_delay ? 1 : 0);
|
|
data_dumper_->DumpRaw("aec3_filter_tail_freq_resp_est",
|
|
GetReverbFrequencyResponse());
|
|
data_dumper_->DumpRaw("aec3_subtractor_y2", subtractor_output[0].y2);
|
|
data_dumper_->DumpRaw("aec3_subtractor_e2_coarse",
|
|
subtractor_output[0].e2_coarse);
|
|
data_dumper_->DumpRaw("aec3_subtractor_e2_refined",
|
|
subtractor_output[0].e2_refined);
|
|
}
|
|
|
|
AecState::InitialState::InitialState(const EchoCanceller3Config& config)
|
|
: conservative_initial_phase_(config.filter.conservative_initial_phase),
|
|
initial_state_seconds_(config.filter.initial_state_seconds) {
|
|
Reset();
|
|
}
|
|
void AecState::InitialState::InitialState::Reset() {
|
|
initial_state_ = true;
|
|
strong_not_saturated_render_blocks_ = 0;
|
|
}
|
|
void AecState::InitialState::InitialState::Update(bool active_render,
|
|
bool saturated_capture) {
|
|
strong_not_saturated_render_blocks_ +=
|
|
active_render && !saturated_capture ? 1 : 0;
|
|
|
|
// Flag whether the initial state is still active.
|
|
bool prev_initial_state = initial_state_;
|
|
if (conservative_initial_phase_) {
|
|
initial_state_ =
|
|
strong_not_saturated_render_blocks_ < 5 * kNumBlocksPerSecond;
|
|
} else {
|
|
initial_state_ = strong_not_saturated_render_blocks_ <
|
|
initial_state_seconds_ * kNumBlocksPerSecond;
|
|
}
|
|
|
|
// Flag whether the transition from the initial state has started.
|
|
transition_triggered_ = !initial_state_ && prev_initial_state;
|
|
}
|
|
|
|
AecState::FilterDelay::FilterDelay(const EchoCanceller3Config& config,
|
|
size_t num_capture_channels)
|
|
: delay_headroom_blocks_(config.delay.delay_headroom_samples / kBlockSize),
|
|
filter_delays_blocks_(num_capture_channels, delay_headroom_blocks_),
|
|
min_filter_delay_(delay_headroom_blocks_) {}
|
|
|
|
void AecState::FilterDelay::Update(
|
|
rtc::ArrayView<const int> analyzer_filter_delay_estimates_blocks,
|
|
const absl::optional<DelayEstimate>& external_delay,
|
|
size_t blocks_with_proper_filter_adaptation) {
|
|
// Update the delay based on the external delay.
|
|
if (external_delay &&
|
|
(!external_delay_ || external_delay_->delay != external_delay->delay)) {
|
|
external_delay_ = external_delay;
|
|
external_delay_reported_ = true;
|
|
}
|
|
|
|
// Override the estimated delay if it is not certain that the filter has had
|
|
// time to converge.
|
|
const bool delay_estimator_may_not_have_converged =
|
|
blocks_with_proper_filter_adaptation < 2 * kNumBlocksPerSecond;
|
|
if (delay_estimator_may_not_have_converged && external_delay_) {
|
|
const int delay_guess = delay_headroom_blocks_;
|
|
std::fill(filter_delays_blocks_.begin(), filter_delays_blocks_.end(),
|
|
delay_guess);
|
|
} else {
|
|
RTC_DCHECK_EQ(filter_delays_blocks_.size(),
|
|
analyzer_filter_delay_estimates_blocks.size());
|
|
std::copy(analyzer_filter_delay_estimates_blocks.begin(),
|
|
analyzer_filter_delay_estimates_blocks.end(),
|
|
filter_delays_blocks_.begin());
|
|
}
|
|
|
|
min_filter_delay_ = *std::min_element(filter_delays_blocks_.begin(),
|
|
filter_delays_blocks_.end());
|
|
}
|
|
|
|
AecState::FilteringQualityAnalyzer::FilteringQualityAnalyzer(
|
|
const EchoCanceller3Config& config,
|
|
size_t num_capture_channels)
|
|
: use_linear_filter_(config.filter.use_linear_filter),
|
|
usable_linear_filter_estimates_(num_capture_channels, false) {}
|
|
|
|
void AecState::FilteringQualityAnalyzer::Reset() {
|
|
std::fill(usable_linear_filter_estimates_.begin(),
|
|
usable_linear_filter_estimates_.end(), false);
|
|
overall_usable_linear_estimates_ = false;
|
|
filter_update_blocks_since_reset_ = 0;
|
|
}
|
|
|
|
void AecState::FilteringQualityAnalyzer::Update(
|
|
bool active_render,
|
|
bool transparent_mode,
|
|
bool saturated_capture,
|
|
const absl::optional<DelayEstimate>& external_delay,
|
|
bool any_filter_converged) {
|
|
// Update blocks counter.
|
|
const bool filter_update = active_render && !saturated_capture;
|
|
filter_update_blocks_since_reset_ += filter_update ? 1 : 0;
|
|
filter_update_blocks_since_start_ += filter_update ? 1 : 0;
|
|
|
|
// Store convergence flag when observed.
|
|
convergence_seen_ = convergence_seen_ || any_filter_converged;
|
|
|
|
// Verify requirements for achieving a decent filter. The requirements for
|
|
// filter adaptation at call startup are more restrictive than after an
|
|
// in-call reset.
|
|
const bool sufficient_data_to_converge_at_startup =
|
|
filter_update_blocks_since_start_ > kNumBlocksPerSecond * 0.4f;
|
|
const bool sufficient_data_to_converge_at_reset =
|
|
sufficient_data_to_converge_at_startup &&
|
|
filter_update_blocks_since_reset_ > kNumBlocksPerSecond * 0.2f;
|
|
|
|
// The linear filter can only be used if it has had time to converge.
|
|
overall_usable_linear_estimates_ = sufficient_data_to_converge_at_startup &&
|
|
sufficient_data_to_converge_at_reset;
|
|
|
|
// The linear filter can only be used if an external delay or convergence have
|
|
// been identified
|
|
overall_usable_linear_estimates_ =
|
|
overall_usable_linear_estimates_ && (external_delay || convergence_seen_);
|
|
|
|
// If transparent mode is on, deactivate usign the linear filter.
|
|
overall_usable_linear_estimates_ =
|
|
overall_usable_linear_estimates_ && !transparent_mode;
|
|
|
|
if (use_linear_filter_) {
|
|
std::fill(usable_linear_filter_estimates_.begin(),
|
|
usable_linear_filter_estimates_.end(),
|
|
overall_usable_linear_estimates_);
|
|
}
|
|
}
|
|
|
|
void AecState::SaturationDetector::Update(
|
|
rtc::ArrayView<const std::vector<float>> x,
|
|
bool saturated_capture,
|
|
bool usable_linear_estimate,
|
|
rtc::ArrayView<const SubtractorOutput> subtractor_output,
|
|
float echo_path_gain) {
|
|
saturated_echo_ = false;
|
|
if (!saturated_capture) {
|
|
return;
|
|
}
|
|
|
|
if (usable_linear_estimate) {
|
|
constexpr float kSaturationThreshold = 20000.f;
|
|
for (size_t ch = 0; ch < subtractor_output.size(); ++ch) {
|
|
saturated_echo_ =
|
|
saturated_echo_ ||
|
|
(subtractor_output[ch].s_refined_max_abs > kSaturationThreshold ||
|
|
subtractor_output[ch].s_coarse_max_abs > kSaturationThreshold);
|
|
}
|
|
} else {
|
|
float max_sample = 0.f;
|
|
for (auto& channel : x) {
|
|
for (float sample : channel) {
|
|
max_sample = std::max(max_sample, fabsf(sample));
|
|
}
|
|
}
|
|
|
|
const float kMargin = 10.f;
|
|
float peak_echo_amplitude = max_sample * echo_path_gain * kMargin;
|
|
saturated_echo_ = saturated_echo_ || peak_echo_amplitude > 32000;
|
|
}
|
|
}
|
|
|
|
} // namespace webrtc
|