Allow extracting the linear AEC output

This CL enables extracting the linear AEC output,
allowing for more straightforward
testing/development.

Bug: b/140823178
Change-Id: I14f7934008d87066b35500466cb6e6d96f811688
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/153672
Commit-Queue: Per Åhgren <peah@webrtc.org>
Reviewed-by: Gustaf Ullberg <gustaf@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#29789}
This commit is contained in:
Per Åhgren 2019-11-13 11:12:29 +01:00 committed by Commit Bot
parent 06a394e036
commit c20a19cc4b
26 changed files with 293 additions and 90 deletions

View file

@ -79,6 +79,7 @@ struct RTC_EXPORT EchoCanceller3Config {
bool conservative_initial_phase = false;
bool enable_shadow_filter_output_usage = true;
bool use_linear_filter = true;
bool export_linear_aec_output = false;
} filter;
struct Erle {
@ -185,8 +186,6 @@ struct RTC_EXPORT EchoCanceller3Config {
} high_bands_suppression;
float floor_first_increase = 0.00001f;
bool enforce_transparent = false;
bool enforce_empty_higher_bands = false;
} suppressor;
};
} // namespace webrtc

View file

@ -190,6 +190,8 @@ void Aec3ConfigFromJsonString(absl::string_view json_string,
ReadParam(section, "enable_shadow_filter_output_usage",
&cfg.filter.enable_shadow_filter_output_usage);
ReadParam(section, "use_linear_filter", &cfg.filter.use_linear_filter);
ReadParam(section, "export_linear_aec_output",
&cfg.filter.export_linear_aec_output);
}
if (rtc::GetValueFromJsonObject(aec3_root, "erle", &section)) {
@ -314,10 +316,6 @@ void Aec3ConfigFromJsonString(absl::string_view json_string,
ReadParam(section, "floor_first_increase",
&cfg.suppressor.floor_first_increase);
ReadParam(section, "enforce_transparent",
&cfg.suppressor.enforce_transparent);
ReadParam(section, "enforce_empty_higher_bands",
&cfg.suppressor.enforce_empty_higher_bands);
}
}
@ -408,7 +406,12 @@ std::string Aec3ConfigToJsonString(const EchoCanceller3Config& config) {
ost << "\"conservative_initial_phase\": "
<< (config.filter.conservative_initial_phase ? "true" : "false") << ",";
ost << "\"enable_shadow_filter_output_usage\": "
<< (config.filter.enable_shadow_filter_output_usage ? "true" : "false");
<< (config.filter.enable_shadow_filter_output_usage ? "true" : "false")
<< ",";
ost << "\"use_linear_filter\": "
<< (config.filter.use_linear_filter ? "true" : "false") << ",";
ost << "\"export_linear_aec_output\": "
<< (config.filter.export_linear_aec_output ? "true" : "false");
ost << "},";
@ -545,12 +548,7 @@ std::string Aec3ConfigToJsonString(const EchoCanceller3Config& config) {
ost << "\"max_gain_during_echo\": "
<< config.suppressor.high_bands_suppression.max_gain_during_echo;
ost << "},";
ost << "\"floor_first_increase\": " << config.suppressor.floor_first_increase
<< ",";
ost << "\"enforce_transparent\": "
<< (config.suppressor.enforce_transparent ? "true" : "false") << ",";
ost << "\"enforce_empty_higher_bands\": "
<< (config.suppressor.enforce_empty_higher_bands ? "true" : "false");
ost << "\"floor_first_increase\": " << config.suppressor.floor_first_increase;
ost << "}";
ost << "}";
ost << "}";

View file

@ -31,6 +31,12 @@ class EchoControl {
// Processes the capture signal in order to remove the echo.
virtual void ProcessCapture(AudioBuffer* capture, bool echo_path_change) = 0;
// As above, but also returns the linear filter output.
// TODO(peah): Make pure virtual.
virtual void ProcessCapture(AudioBuffer* capture,
AudioBuffer* linear_output,
bool level_change) {}
struct Metrics {
double echo_return_loss;
double echo_return_loss_enhancement;

View file

@ -42,7 +42,8 @@ constexpr int kMaxAdaptiveFilterLength = 50;
constexpr int kRenderTransferQueueSizeFrames = 100;
constexpr size_t kMaxNumBands = 3;
constexpr size_t kSubFrameLength = 80;
constexpr size_t kFrameSize = 160;
constexpr size_t kSubFrameLength = kFrameSize / 2;
constexpr size_t kBlockSize = kFftLengthBy2;
constexpr size_t kBlockSizeLog2 = kFftLengthBy2Log2;

View file

@ -52,6 +52,7 @@ class BlockProcessorImpl final : public BlockProcessor {
void ProcessCapture(
bool echo_path_gain_change,
bool capture_signal_saturation,
std::vector<std::vector<std::vector<float>>>* linear_output,
std::vector<std::vector<std::vector<float>>>* capture_block) override;
void BufferRender(
@ -105,6 +106,7 @@ BlockProcessorImpl::~BlockProcessorImpl() = default;
void BlockProcessorImpl::ProcessCapture(
bool echo_path_gain_change,
bool capture_signal_saturation,
std::vector<std::vector<std::vector<float>>>* linear_output,
std::vector<std::vector<std::vector<float>>>* capture_block) {
RTC_DCHECK(capture_block);
RTC_DCHECK_EQ(NumBandsForRate(sample_rate_hz_), capture_block->size());
@ -191,7 +193,7 @@ void BlockProcessorImpl::ProcessCapture(
if (has_delay_estimator || render_buffer_->HasReceivedBufferDelay()) {
echo_remover_->ProcessCapture(
echo_path_variability, capture_signal_saturation, estimated_delay_,
render_buffer_->GetRenderBuffer(), capture_block);
render_buffer_->GetRenderBuffer(), linear_output, capture_block);
}
// Update the metrics.

View file

@ -59,6 +59,7 @@ class BlockProcessor {
virtual void ProcessCapture(
bool echo_path_gain_change,
bool capture_signal_saturation,
std::vector<std::vector<std::vector<float>>>* linear_output,
std::vector<std::vector<std::vector<float>>>* capture_block) = 0;
// Buffers a block of render data supplied by a FrameBlocker object.

View file

@ -48,7 +48,7 @@ void RunBasicSetupAndApiCallTest(int sample_rate_hz, int num_iterations) {
std::vector<float>(kBlockSize, 1000.f)));
for (int k = 0; k < num_iterations; ++k) {
block_processor->BufferRender(block);
block_processor->ProcessCapture(false, false, &block);
block_processor->ProcessCapture(false, false, nullptr, &block);
block_processor->UpdateEchoLeakageStatus(false);
}
}
@ -81,7 +81,8 @@ void RunCaptureBlockSizeVerificationTest(int sample_rate_hz) {
std::vector<std::vector<float>>(kNumRenderChannels,
std::vector<float>(kBlockSize - 1, 0.f)));
EXPECT_DEATH(block_processor->ProcessCapture(false, false, &block), "");
EXPECT_DEATH(block_processor->ProcessCapture(false, false, nullptr, &block),
"");
}
void RunRenderNumBandsVerificationTest(int sample_rate_hz) {
@ -117,7 +118,8 @@ void RunCaptureNumBandsVerificationTest(int sample_rate_hz) {
std::vector<std::vector<float>>(kNumRenderChannels,
std::vector<float>(kBlockSize, 0.f)));
EXPECT_DEATH(block_processor->ProcessCapture(false, false, &block), "");
EXPECT_DEATH(block_processor->ProcessCapture(false, false, nullptr, &block),
"");
}
#endif
@ -172,7 +174,7 @@ TEST(BlockProcessor, DISABLED_DelayControllerIntegration) {
RandomizeSampleVector(&random_generator, render_block[0][0]);
signal_delay_buffer.Delay(render_block[0][0], capture_block[0][0]);
block_processor->BufferRender(render_block);
block_processor->ProcessCapture(false, false, &capture_block);
block_processor->ProcessCapture(false, false, nullptr, &capture_block);
}
}
}
@ -207,7 +209,7 @@ TEST(BlockProcessor, DISABLED_SubmoduleIntegration) {
.WillRepeatedly(Return(0));
EXPECT_CALL(*render_delay_controller_mock, GetDelay(_, _, _))
.Times(kNumBlocks);
EXPECT_CALL(*echo_remover_mock, ProcessCapture(_, _, _, _, _))
EXPECT_CALL(*echo_remover_mock, ProcessCapture(_, _, _, _, _, _))
.Times(kNumBlocks);
EXPECT_CALL(*echo_remover_mock, UpdateEchoLeakageStatus(_))
.Times(kNumBlocks);
@ -230,7 +232,7 @@ TEST(BlockProcessor, DISABLED_SubmoduleIntegration) {
RandomizeSampleVector(&random_generator, render_block[0][0]);
signal_delay_buffer.Delay(render_block[0][0], capture_block[0][0]);
block_processor->BufferRender(render_block);
block_processor->ProcessCapture(false, false, &capture_block);
block_processor->ProcessCapture(false, false, nullptr, &capture_block);
block_processor->UpdateEchoLeakageStatus(false);
}
}
@ -284,7 +286,7 @@ TEST(BlockProcessor, VerifyCaptureNumBandsCheck) {
TEST(BlockProcessor, NullProcessCaptureParameter) {
EXPECT_DEATH(std::unique_ptr<BlockProcessor>(
BlockProcessor::Create(EchoCanceller3Config(), 16000, 1, 1))
->ProcessCapture(false, false, nullptr),
->ProcessCapture(false, false, nullptr, nullptr),
"");
}

View file

@ -16,6 +16,7 @@
#include "modules/audio_processing/high_pass_filter.h"
#include "modules/audio_processing/logging/apm_data_dumper.h"
#include "rtc_base/atomic_ops.h"
#include "rtc_base/logging.h"
#include "system_wrappers/include/field_trial.h"
namespace webrtc {
@ -87,28 +88,52 @@ void FillSubFrameView(
}
void ProcessCaptureFrameContent(
AudioBuffer* linear_output,
AudioBuffer* capture,
bool level_change,
bool saturated_microphone_signal,
size_t sub_frame_index,
FrameBlocker* capture_blocker,
BlockFramer* linear_output_framer,
BlockFramer* output_framer,
BlockProcessor* block_processor,
std::vector<std::vector<std::vector<float>>>* block,
std::vector<std::vector<rtc::ArrayView<float>>>* sub_frame_view) {
FillSubFrameView(capture, sub_frame_index, sub_frame_view);
capture_blocker->InsertSubFrameAndExtractBlock(*sub_frame_view, block);
std::vector<std::vector<std::vector<float>>>* linear_output_block,
std::vector<std::vector<rtc::ArrayView<float>>>*
linear_output_sub_frame_view,
std::vector<std::vector<std::vector<float>>>* capture_block,
std::vector<std::vector<rtc::ArrayView<float>>>* capture_sub_frame_view) {
FillSubFrameView(capture, sub_frame_index, capture_sub_frame_view);
if (linear_output) {
RTC_DCHECK(linear_output_framer);
RTC_DCHECK(linear_output_block);
RTC_DCHECK(linear_output_sub_frame_view);
FillSubFrameView(linear_output, sub_frame_index,
linear_output_sub_frame_view);
}
capture_blocker->InsertSubFrameAndExtractBlock(*capture_sub_frame_view,
capture_block);
block_processor->ProcessCapture(level_change, saturated_microphone_signal,
block);
output_framer->InsertBlockAndExtractSubFrame(*block, sub_frame_view);
linear_output_block, capture_block);
output_framer->InsertBlockAndExtractSubFrame(*capture_block,
capture_sub_frame_view);
if (linear_output) {
RTC_DCHECK(linear_output_framer);
linear_output_framer->InsertBlockAndExtractSubFrame(
*linear_output_block, linear_output_sub_frame_view);
}
}
void ProcessRemainingCaptureFrameContent(
bool level_change,
bool saturated_microphone_signal,
FrameBlocker* capture_blocker,
BlockFramer* linear_output_framer,
BlockFramer* output_framer,
BlockProcessor* block_processor,
std::vector<std::vector<std::vector<float>>>* linear_output_block,
std::vector<std::vector<std::vector<float>>>* block) {
if (!capture_blocker->IsBlockAvailable()) {
return;
@ -116,8 +141,13 @@ void ProcessRemainingCaptureFrameContent(
capture_blocker->ExtractBlock(block);
block_processor->ProcessCapture(level_change, saturated_microphone_signal,
block);
linear_output_block, block);
output_framer->InsertBlock(*block);
if (linear_output_framer) {
RTC_DCHECK(linear_output_block);
linear_output_framer->InsertBlock(*linear_output_block);
}
}
void BufferRenderFrameContent(
@ -295,12 +325,24 @@ EchoCanceller3::EchoCanceller3(const EchoCanceller3Config& config,
RTC_DCHECK_EQ(num_bands_, std::max(sample_rate_hz_, 16000) / 16000);
RTC_DCHECK_GE(kMaxNumBands, num_bands_);
if (config_.filter.export_linear_aec_output) {
linear_output_framer_.reset(new BlockFramer(1, num_capture_channels_));
linear_output_block_ =
std::make_unique<std::vector<std::vector<std::vector<float>>>>(
1, std::vector<std::vector<float>>(
num_capture_channels_, std::vector<float>(kBlockSize, 0.f)));
linear_output_sub_frame_view_ =
std::vector<std::vector<rtc::ArrayView<float>>>(
1, std::vector<rtc::ArrayView<float>>(num_capture_channels_));
}
}
EchoCanceller3::~EchoCanceller3() = default;
void EchoCanceller3::AnalyzeRender(const AudioBuffer& render) {
RTC_DCHECK_RUNS_SERIALIZED(&render_race_checker_);
RTC_DCHECK_EQ(render.num_channels(), num_render_channels_);
data_dumper_->DumpRaw("aec3_call_order",
static_cast<int>(EchoCanceller3ApiCall::kRender));
@ -312,7 +354,6 @@ void EchoCanceller3::AnalyzeCapture(const AudioBuffer& capture) {
RTC_DCHECK_RUNS_SERIALIZED(&capture_race_checker_);
data_dumper_->DumpWav("aec3_capture_analyze_input", capture.num_frames(),
capture.channels_const()[0], sample_rate_hz_, 1);
saturated_microphone_signal_ = false;
for (size_t channel = 0; channel < capture.num_channels(); ++channel) {
saturated_microphone_signal_ |=
@ -325,6 +366,12 @@ void EchoCanceller3::AnalyzeCapture(const AudioBuffer& capture) {
}
void EchoCanceller3::ProcessCapture(AudioBuffer* capture, bool level_change) {
ProcessCapture(capture, nullptr, level_change);
}
void EchoCanceller3::ProcessCapture(AudioBuffer* capture,
AudioBuffer* linear_output,
bool level_change) {
RTC_DCHECK_RUNS_SERIALIZED(&capture_race_checker_);
RTC_DCHECK(capture);
RTC_DCHECK_EQ(num_bands_, capture->num_bands());
@ -333,6 +380,12 @@ void EchoCanceller3::ProcessCapture(AudioBuffer* capture, bool level_change) {
data_dumper_->DumpRaw("aec3_call_order",
static_cast<int>(EchoCanceller3ApiCall::kCapture));
if (linear_output && !linear_output_framer_) {
RTC_LOG(LS_ERROR) << "Trying to retrieve the linear AEC output without "
"properly configuring AEC3.";
RTC_NOTREACHED();
}
// Report capture call in the metrics and periodically update API call
// metrics.
api_call_metrics_.ReportCaptureCall();
@ -349,19 +402,24 @@ void EchoCanceller3::ProcessCapture(AudioBuffer* capture, bool level_change) {
EmptyRenderQueue();
ProcessCaptureFrameContent(capture, level_change,
ProcessCaptureFrameContent(linear_output, capture, level_change,
saturated_microphone_signal_, 0, &capture_blocker_,
&output_framer_, block_processor_.get(),
&capture_block_, &capture_sub_frame_view_);
linear_output_framer_.get(), &output_framer_,
block_processor_.get(), linear_output_block_.get(),
&linear_output_sub_frame_view_, &capture_block_,
&capture_sub_frame_view_);
ProcessCaptureFrameContent(capture, level_change,
ProcessCaptureFrameContent(linear_output, capture, level_change,
saturated_microphone_signal_, 1, &capture_blocker_,
&output_framer_, block_processor_.get(),
&capture_block_, &capture_sub_frame_view_);
linear_output_framer_.get(), &output_framer_,
block_processor_.get(), linear_output_block_.get(),
&linear_output_sub_frame_view_, &capture_block_,
&capture_sub_frame_view_);
ProcessRemainingCaptureFrameContent(
level_change, saturated_microphone_signal_, &capture_blocker_,
&output_framer_, block_processor_.get(), &capture_block_);
linear_output_framer_.get(), &output_framer_, block_processor_.get(),
linear_output_block_.get(), &capture_block_);
data_dumper_->DumpWav("aec3_capture_output", AudioBuffer::kSplitBandSize,
&capture->split_bands(0)[0][0], 16000, 1);

View file

@ -70,8 +70,6 @@ class Aec3RenderQueueItemVerifier {
// Main class for the echo canceller3.
// It does 4 things:
// -Receives 10 ms frames of band-split audio.
// -Optionally applies an anti-hum (high-pass) filter on the
// received signals.
// -Provides the lower level echo canceller functionality with
// blocks of 64 samples of audio data.
// -Partially handles the jitter in the render and capture API
@ -106,6 +104,10 @@ class EchoCanceller3 : public EchoControl {
// Processes the split-band domain capture signal in order to remove any echo
// present in the signal.
void ProcessCapture(AudioBuffer* capture, bool level_change) override;
// As above, but also returns the linear filter output.
void ProcessCapture(AudioBuffer* capture,
AudioBuffer* linear_output,
bool level_change) override;
// Collect current metrics from the echo canceller.
Metrics GetMetrics() const override;
// Provides an optional external estimate of the audio buffer delay.
@ -149,6 +151,8 @@ class EchoCanceller3 : public EchoControl {
const int num_bands_;
const size_t num_render_channels_;
const size_t num_capture_channels_;
std::unique_ptr<BlockFramer> linear_output_framer_
RTC_GUARDED_BY(capture_race_checker_);
BlockFramer output_framer_ RTC_GUARDED_BY(capture_race_checker_);
FrameBlocker capture_blocker_ RTC_GUARDED_BY(capture_race_checker_);
FrameBlocker render_blocker_ RTC_GUARDED_BY(capture_race_checker_);
@ -163,10 +167,14 @@ class EchoCanceller3 : public EchoControl {
false;
std::vector<std::vector<std::vector<float>>> render_block_
RTC_GUARDED_BY(capture_race_checker_);
std::unique_ptr<std::vector<std::vector<std::vector<float>>>>
linear_output_block_ RTC_GUARDED_BY(capture_race_checker_);
std::vector<std::vector<std::vector<float>>> capture_block_
RTC_GUARDED_BY(capture_race_checker_);
std::vector<std::vector<rtc::ArrayView<float>>> render_sub_frame_view_
RTC_GUARDED_BY(capture_race_checker_);
std::vector<std::vector<rtc::ArrayView<float>>> linear_output_sub_frame_view_
RTC_GUARDED_BY(capture_race_checker_);
std::vector<std::vector<rtc::ArrayView<float>>> capture_sub_frame_view_
RTC_GUARDED_BY(capture_race_checker_);
BlockDelayBuffer block_delay_buffer_ RTC_GUARDED_BY(capture_race_checker_);

View file

@ -112,6 +112,7 @@ class CaptureTransportVerificationProcessor : public BlockProcessor {
void ProcessCapture(
bool level_change,
bool saturated_microphone_signal,
std::vector<std::vector<std::vector<float>>>* linear_output,
std::vector<std::vector<std::vector<float>>>* capture_block) override {}
void BufferRender(
@ -137,6 +138,7 @@ class RenderTransportVerificationProcessor : public BlockProcessor {
void ProcessCapture(
bool level_change,
bool saturated_microphone_signal,
std::vector<std::vector<std::vector<float>>>* linear_output,
std::vector<std::vector<std::vector<float>>>* capture_block) override {
std::vector<std::vector<std::vector<float>>> render_block =
received_render_blocks_.front();
@ -267,17 +269,17 @@ class EchoCanceller3Tester {
switch (echo_path_change_test_variant) {
case EchoPathChangeTestVariant::kNone:
EXPECT_CALL(*block_processor_mock, ProcessCapture(false, _, _))
EXPECT_CALL(*block_processor_mock, ProcessCapture(false, _, _, _))
.Times(kExpectedNumBlocksToProcess);
break;
case EchoPathChangeTestVariant::kOneSticky:
EXPECT_CALL(*block_processor_mock, ProcessCapture(true, _, _))
EXPECT_CALL(*block_processor_mock, ProcessCapture(true, _, _, _))
.Times(kExpectedNumBlocksToProcess);
break;
case EchoPathChangeTestVariant::kOneNonSticky:
EXPECT_CALL(*block_processor_mock, ProcessCapture(true, _, _))
EXPECT_CALL(*block_processor_mock, ProcessCapture(true, _, _, _))
.Times(kNumFullBlocksPerFrame);
EXPECT_CALL(*block_processor_mock, ProcessCapture(false, _, _))
EXPECT_CALL(*block_processor_mock, ProcessCapture(false, _, _, _))
.Times(kExpectedNumBlocksToProcess - kNumFullBlocksPerFrame);
break;
}
@ -338,7 +340,7 @@ class EchoCanceller3Tester {
new StrictMock<webrtc::test::MockBlockProcessor>());
EXPECT_CALL(*block_processor_mock, BufferRender(_))
.Times(kExpectedNumBlocksToProcess);
EXPECT_CALL(*block_processor_mock, ProcessCapture(_, _, _))
EXPECT_CALL(*block_processor_mock, ProcessCapture(_, _, _, _))
.Times(kExpectedNumBlocksToProcess);
switch (leakage_report_variant) {
@ -429,21 +431,21 @@ class EchoCanceller3Tester {
switch (saturation_variant) {
case SaturationTestVariant::kNone:
EXPECT_CALL(*block_processor_mock, ProcessCapture(_, false, _))
EXPECT_CALL(*block_processor_mock, ProcessCapture(_, false, _, _))
.Times(kExpectedNumBlocksToProcess);
break;
case SaturationTestVariant::kOneNegative: {
::testing::InSequence s;
EXPECT_CALL(*block_processor_mock, ProcessCapture(_, true, _))
EXPECT_CALL(*block_processor_mock, ProcessCapture(_, true, _, _))
.Times(kNumFullBlocksPerFrame);
EXPECT_CALL(*block_processor_mock, ProcessCapture(_, false, _))
EXPECT_CALL(*block_processor_mock, ProcessCapture(_, false, _, _))
.Times(kExpectedNumBlocksToProcess - kNumFullBlocksPerFrame);
} break;
case SaturationTestVariant::kOnePositive: {
::testing::InSequence s;
EXPECT_CALL(*block_processor_mock, ProcessCapture(_, true, _))
EXPECT_CALL(*block_processor_mock, ProcessCapture(_, true, _, _))
.Times(kNumFullBlocksPerFrame);
EXPECT_CALL(*block_processor_mock, ProcessCapture(_, false, _))
EXPECT_CALL(*block_processor_mock, ProcessCapture(_, false, _, _))
.Times(kExpectedNumBlocksToProcess - kNumFullBlocksPerFrame);
} break;
}

View file

@ -123,6 +123,7 @@ class EchoRemoverImpl final : public EchoRemover {
bool capture_signal_saturation,
const absl::optional<DelayEstimate>& external_delay,
RenderBuffer* render_buffer,
std::vector<std::vector<std::vector<float>>>* linear_output,
std::vector<std::vector<std::vector<float>>>* capture) override;
// Updates the status on whether echo leakage is detected in the output of the
@ -235,6 +236,7 @@ void EchoRemoverImpl::ProcessCapture(
bool capture_signal_saturation,
const absl::optional<DelayEstimate>& external_delay,
RenderBuffer* render_buffer,
std::vector<std::vector<std::vector<float>>>* linear_output,
std::vector<std::vector<std::vector<float>>>* capture) {
++block_counter_;
const std::vector<std::vector<std::vector<float>>>& x =
@ -367,6 +369,16 @@ void EchoRemoverImpl::ProcessCapture(
E[ch].Spectrum(optimization_, E2[ch]);
}
// Optionally return the linear filter output.
if (linear_output) {
RTC_DCHECK_GE(1, linear_output->size());
RTC_DCHECK_EQ(num_capture_channels_, linear_output[0].size());
for (size_t ch = 0; ch < num_capture_channels_; ++ch) {
RTC_DCHECK_EQ(kBlockSize, (*linear_output)[0][ch].size());
std::copy(e[ch].begin(), e[ch].end(), (*linear_output)[0][ch].begin());
}
}
// Update the AEC state information.
aec_state_.Update(external_delay, subtractor_.FilterFrequencyResponses(),
subtractor_.FilterImpulseResponses(), *render_buffer, E2,

View file

@ -42,6 +42,7 @@ class EchoRemover {
bool capture_signal_saturation,
const absl::optional<DelayEstimate>& external_delay,
RenderBuffer* render_buffer,
std::vector<std::vector<std::vector<float>>>* linear_output,
std::vector<std::vector<std::vector<float>>>* capture) = 0;
// Updates the status on whether echo leakage is detected in the output of the

View file

@ -73,9 +73,9 @@ TEST(EchoRemover, BasicApiCalls) {
render_buffer->Insert(render);
render_buffer->PrepareCaptureProcessing();
remover->ProcessCapture(echo_path_variability,
k % 2 == 0 ? true : false, delay_estimate,
render_buffer->GetRenderBuffer(), &capture);
remover->ProcessCapture(
echo_path_variability, k % 2 == 0 ? true : false, delay_estimate,
render_buffer->GetRenderBuffer(), nullptr, &capture);
}
}
}
@ -107,9 +107,9 @@ TEST(EchoRemover, WrongCaptureBlockSize) {
1, std::vector<float>(kBlockSize - 1, 0.f)));
EchoPathVariability echo_path_variability(
false, EchoPathVariability::DelayAdjustment::kNone, false);
EXPECT_DEATH(
remover->ProcessCapture(echo_path_variability, false, delay_estimate,
render_buffer->GetRenderBuffer(), &capture),
EXPECT_DEATH(remover->ProcessCapture(
echo_path_variability, false, delay_estimate,
render_buffer->GetRenderBuffer(), nullptr, &capture),
"");
}
}
@ -131,9 +131,9 @@ TEST(EchoRemover, DISABLED_WrongCaptureNumBands) {
std::vector<float>(kBlockSize, 0.f)));
EchoPathVariability echo_path_variability(
false, EchoPathVariability::DelayAdjustment::kNone, false);
EXPECT_DEATH(
remover->ProcessCapture(echo_path_variability, false, delay_estimate,
render_buffer->GetRenderBuffer(), &capture),
EXPECT_DEATH(remover->ProcessCapture(
echo_path_variability, false, delay_estimate,
render_buffer->GetRenderBuffer(), nullptr, &capture),
"");
}
}
@ -147,9 +147,9 @@ TEST(EchoRemover, NullCapture) {
RenderDelayBuffer::Create(EchoCanceller3Config(), 16000, 1));
EchoPathVariability echo_path_variability(
false, EchoPathVariability::DelayAdjustment::kNone, false);
EXPECT_DEATH(
remover->ProcessCapture(echo_path_variability, false, delay_estimate,
render_buffer->GetRenderBuffer(), nullptr),
EXPECT_DEATH(remover->ProcessCapture(
echo_path_variability, false, delay_estimate,
render_buffer->GetRenderBuffer(), nullptr, nullptr),
"");
}
@ -222,7 +222,8 @@ TEST(EchoRemover, BasicEchoRemoval) {
render_buffer->PrepareCaptureProcessing();
remover->ProcessCapture(echo_path_variability, false, delay_estimate,
render_buffer->GetRenderBuffer(), &y);
render_buffer->GetRenderBuffer(), nullptr,
&y);
if (k > kNumBlocksToProcess / 2) {
output_energy = std::inner_product(y[0][0].begin(), y[0][0].end(),

View file

@ -24,10 +24,11 @@ class MockBlockProcessor : public BlockProcessor {
MockBlockProcessor();
virtual ~MockBlockProcessor();
MOCK_METHOD3(
MOCK_METHOD4(
ProcessCapture,
void(bool level_change,
bool saturated_microphone_signal,
std::vector<std::vector<std::vector<float>>>* linear_output,
std::vector<std::vector<std::vector<float>>>* capture_block));
MOCK_METHOD1(BufferRender,
void(const std::vector<std::vector<std::vector<float>>>& block));

View file

@ -27,11 +27,12 @@ class MockEchoRemover : public EchoRemover {
MockEchoRemover();
virtual ~MockEchoRemover();
MOCK_METHOD5(ProcessCapture,
MOCK_METHOD6(ProcessCapture,
void(EchoPathVariability echo_path_variability,
bool capture_signal_saturation,
const absl::optional<DelayEstimate>& delay_estimate,
RenderBuffer* render_buffer,
std::vector<std::vector<std::vector<float>>>* linear_output,
std::vector<std::vector<std::vector<float>>>* capture));
MOCK_CONST_METHOD0(Delay, absl::optional<int>());
MOCK_METHOD1(UpdateEchoLeakageStatus, void(bool leakage_detected));

View file

@ -343,13 +343,6 @@ void SuppressionGain::GetGain(
std::array<float, kFftLengthBy2Plus1>* low_band_gain) {
RTC_DCHECK(high_bands_gain);
RTC_DCHECK(low_band_gain);
const auto& cfg = config_.suppressor;
if (cfg.enforce_transparent) {
low_band_gain->fill(1.f);
*high_bands_gain = cfg.enforce_empty_higher_bands ? 0.f : 1.f;
return;
}
// Update the nearend state selection.
dominant_nearend_detector_.Update(nearend_spectrum, residual_echo_spectrum,
@ -360,11 +353,6 @@ void SuppressionGain::GetGain(
LowerBandGain(low_noise_render, aec_state, nearend_spectrum,
residual_echo_spectrum, comfort_noise_spectrum, low_band_gain);
if (cfg.enforce_empty_higher_bands) {
*high_bands_gain = 0.f;
return;
}
// Compute the gain for the upper bands.
const absl::optional<int> narrow_peak_band =
render_signal_analyzer.NarrowPeakBand();

View file

@ -1342,8 +1342,9 @@ int AudioProcessingImpl::ProcessCaptureStreamLocked() {
submodules_.echo_controller->SetAudioBufferDelay(stream_delay_ms());
}
AudioBuffer* linear_aec_buffer = capture_.linear_aec_output.get();
submodules_.echo_controller->ProcessCapture(
capture_buffer, capture_.echo_path_gain_change);
capture_buffer, linear_aec_buffer, capture_.echo_path_gain_change);
} else if (submodules_.echo_cancellation) {
// Ensure that the stream delay was set before the call to the
// AEC ProcessCaptureAudio function.
@ -1625,6 +1626,31 @@ int AudioProcessingImpl::set_stream_delay_ms(int delay) {
return retval;
}
bool AudioProcessingImpl::GetLinearAecOutput(
rtc::ArrayView<std::array<float, 160>> linear_output) const {
rtc::CritScope cs(&crit_capture_);
AudioBuffer* linear_aec_buffer = capture_.linear_aec_output.get();
RTC_DCHECK(linear_aec_buffer);
if (linear_aec_buffer) {
RTC_DCHECK_EQ(1, linear_aec_buffer->num_bands());
RTC_DCHECK_EQ(linear_output.size(), linear_aec_buffer->num_channels());
for (size_t ch = 0; ch < linear_aec_buffer->num_channels(); ++ch) {
RTC_DCHECK_EQ(linear_output[ch].size(), linear_aec_buffer->num_frames());
rtc::ArrayView<const float> channel_view =
rtc::ArrayView<const float>(linear_aec_buffer->channels_const()[ch],
linear_aec_buffer->num_frames());
std::copy(channel_view.begin(), channel_view.end(),
linear_output[ch].begin());
}
return true;
}
RTC_LOG(LS_ERROR) << "No linear AEC output available";
RTC_NOTREACHED();
return false;
}
int AudioProcessingImpl::stream_delay_ms() const {
// Used as callback from submodules, hence locking is not allowed.
return capture_nonlocked_.stream_delay_ms;
@ -1790,6 +1816,16 @@ void AudioProcessingImpl::InitializeEchoController() {
num_proc_channels());
}
// Setup the storage for returning the linear AEC output.
if (config_.echo_canceller.export_linear_aec_output) {
constexpr int kLinearOutputRateHz = 16000;
capture_.linear_aec_output = std::make_unique<AudioBuffer>(
kLinearOutputRateHz, num_proc_channels(), kLinearOutputRateHz,
num_proc_channels(), kLinearOutputRateHz, num_proc_channels());
} else {
capture_.linear_aec_output.reset();
}
capture_nonlocked_.echo_controller_enabled = true;
submodules_.echo_cancellation.reset();
@ -1801,6 +1837,7 @@ void AudioProcessingImpl::InitializeEchoController() {
submodules_.echo_controller.reset();
capture_nonlocked_.echo_controller_enabled = false;
capture_.linear_aec_output.reset();
if (!config_.echo_canceller.enabled) {
submodules_.echo_cancellation.reset();

View file

@ -87,6 +87,8 @@ class AudioProcessingImpl : public AudioProcessing {
const StreamConfig& input_config,
const StreamConfig& output_config,
float* const* dest) override;
bool GetLinearAecOutput(
rtc::ArrayView<std::array<float, 160>> linear_output) const override;
void set_output_will_be_muted(bool muted) override;
int set_stream_delay_ms(int delay) override;
void set_delay_offset_ms(int offset) override;
@ -412,6 +414,7 @@ class AudioProcessingImpl : public AudioProcessing {
bool transient_suppressor_enabled;
std::unique_ptr<AudioBuffer> capture_audio;
std::unique_ptr<AudioBuffer> capture_fullband_audio;
std::unique_ptr<AudioBuffer> linear_aec_output;
// Only the rate and samples fields of capture_processing_format_ are used
// because the capture processing number of channels is mutable and is
// tracked by the capture_audio_.

View file

@ -242,13 +242,13 @@ TEST(AudioProcessingImplTest,
EXPECT_CALL(*echo_control_mock, AnalyzeCapture(testing::_)).Times(1);
EXPECT_CALL(*echo_control_mock,
ProcessCapture(NotNull(), /*echo_path_change=*/false))
ProcessCapture(NotNull(), testing::_, /*echo_path_change=*/false))
.Times(1);
apm->ProcessStream(&frame);
EXPECT_CALL(*echo_control_mock, AnalyzeCapture(testing::_)).Times(1);
EXPECT_CALL(*echo_control_mock,
ProcessCapture(NotNull(), /*echo_path_change=*/true))
ProcessCapture(NotNull(), testing::_, /*echo_path_change=*/true))
.Times(1);
apm->SetRuntimeSetting(
AudioProcessing::RuntimeSetting::CreateCapturePreGain(2.f));
@ -286,7 +286,8 @@ TEST(AudioProcessingImplTest,
const int initial_analog_gain = apm->recommended_stream_analog_level();
EXPECT_CALL(*echo_control_mock, AnalyzeCapture(testing::_)).Times(1);
EXPECT_CALL(*echo_control_mock, ProcessCapture(NotNull(), false)).Times(1);
EXPECT_CALL(*echo_control_mock, ProcessCapture(NotNull(), testing::_, false))
.Times(1);
apm->ProcessStream(&frame);
// Force an analog gain change if it did not happen.
@ -295,7 +296,8 @@ TEST(AudioProcessingImplTest,
}
EXPECT_CALL(*echo_control_mock, AnalyzeCapture(testing::_)).Times(1);
EXPECT_CALL(*echo_control_mock, ProcessCapture(NotNull(), true)).Times(1);
EXPECT_CALL(*echo_control_mock, ProcessCapture(NotNull(), testing::_, true))
.Times(1);
apm->ProcessStream(&frame);
}
@ -326,13 +328,13 @@ TEST(AudioProcessingImplTest, EchoControllerObservesPlayoutVolumeChange) {
EXPECT_CALL(*echo_control_mock, AnalyzeCapture(testing::_)).Times(1);
EXPECT_CALL(*echo_control_mock,
ProcessCapture(NotNull(), /*echo_path_change=*/false))
ProcessCapture(NotNull(), testing::_, /*echo_path_change=*/false))
.Times(1);
apm->ProcessStream(&frame);
EXPECT_CALL(*echo_control_mock, AnalyzeCapture(testing::_)).Times(1);
EXPECT_CALL(*echo_control_mock,
ProcessCapture(NotNull(), /*echo_path_change=*/false))
ProcessCapture(NotNull(), testing::_, /*echo_path_change=*/false))
.Times(1);
apm->SetRuntimeSetting(
AudioProcessing::RuntimeSetting::CreatePlayoutVolumeChange(50));
@ -340,7 +342,7 @@ TEST(AudioProcessingImplTest, EchoControllerObservesPlayoutVolumeChange) {
EXPECT_CALL(*echo_control_mock, AnalyzeCapture(testing::_)).Times(1);
EXPECT_CALL(*echo_control_mock,
ProcessCapture(NotNull(), /*echo_path_change=*/false))
ProcessCapture(NotNull(), testing::_, /*echo_path_change=*/false))
.Times(1);
apm->SetRuntimeSetting(
AudioProcessing::RuntimeSetting::CreatePlayoutVolumeChange(50));
@ -348,7 +350,7 @@ TEST(AudioProcessingImplTest, EchoControllerObservesPlayoutVolumeChange) {
EXPECT_CALL(*echo_control_mock, AnalyzeCapture(testing::_)).Times(1);
EXPECT_CALL(*echo_control_mock,
ProcessCapture(NotNull(), /*echo_path_change=*/true))
ProcessCapture(NotNull(), testing::_, /*echo_path_change=*/true))
.Times(1);
apm->SetRuntimeSetting(
AudioProcessing::RuntimeSetting::CreatePlayoutVolumeChange(100));

View file

@ -2425,7 +2425,8 @@ class MyEchoControlFactory : public EchoControlFactory {
auto ec = new test::MockEchoControl();
EXPECT_CALL(*ec, AnalyzeRender(::testing::_)).Times(1);
EXPECT_CALL(*ec, AnalyzeCapture(::testing::_)).Times(2);
EXPECT_CALL(*ec, ProcessCapture(::testing::_, ::testing::_)).Times(2);
EXPECT_CALL(*ec, ProcessCapture(::testing::_, ::testing::_, ::testing::_))
.Times(2);
return std::unique_ptr<EchoControl>(ec);
}

View file

@ -280,6 +280,7 @@ class RTC_EXPORT AudioProcessing : public rtc::RefCountInterface {
bool legacy_moderate_suppression_level = false;
// Recommended not to use. Will be removed in the future.
bool use_legacy_aec = false;
bool export_linear_aec_output = false;
} echo_canceller;
// Enables background noise suppression.
@ -611,6 +612,13 @@ class RTC_EXPORT AudioProcessing : public rtc::RefCountInterface {
virtual int AnalyzeReverseStream(const float* const* data,
const StreamConfig& reverse_config) = 0;
// Returns the most recently produced 10 ms of the linear AEC output at a rate
// of 16 kHz. If there is more than one capture channel, a mono representation
// of the input is returned. Returns true/false to indicate whether an output
// returned.
virtual bool GetLinearAecOutput(
rtc::ArrayView<std::array<float, 160>> linear_output) const = 0;
// This must be called prior to ProcessStream() if and only if adaptive analog
// gain control is enabled, to pass the current analog level from the audio
// HAL. Must be within the range provided in Config::GainController1.

View file

@ -47,6 +47,10 @@ class MockEchoControl : public EchoControl {
MOCK_METHOD1(AnalyzeCapture, void(AudioBuffer* capture));
MOCK_METHOD2(ProcessCapture,
void(AudioBuffer* capture, bool echo_path_change));
MOCK_METHOD3(ProcessCapture,
void(AudioBuffer* capture,
AudioBuffer* linear_output,
bool echo_path_change));
MOCK_CONST_METHOD0(GetMetrics, Metrics());
MOCK_METHOD1(SetAudioBufferDelay, void(int delay_ms));
MOCK_CONST_METHOD0(ActiveProcessing, bool());
@ -105,6 +109,9 @@ class MockAudioProcessing : public ::testing::NiceMock<AudioProcessing> {
const StreamConfig& input_config,
const StreamConfig& output_config,
float* const* dest));
MOCK_CONST_METHOD1(
GetLinearAecOutput,
bool(rtc::ArrayView<std::array<float, 160>> linear_output));
MOCK_METHOD1(set_stream_delay_ms, int(int delay));
MOCK_CONST_METHOD0(stream_delay_ms, int());
MOCK_CONST_METHOD0(was_stream_delay_set, bool());

View file

@ -227,6 +227,20 @@ void AudioProcessingSimulator::ProcessStream(bool fixed_interface) {
buffer_file_writer_->Write(*out_buf_);
}
if (linear_aec_output_file_writer_) {
bool output_available = ap_->GetLinearAecOutput(linear_aec_output_buf_);
RTC_CHECK(output_available);
RTC_CHECK_GT(linear_aec_output_buf_.size(), 0);
RTC_CHECK_EQ(linear_aec_output_buf_[0].size(), 160);
for (size_t k = 0; k < linear_aec_output_buf_[0].size(); ++k) {
for (size_t ch = 0; ch < linear_aec_output_buf_.size(); ++ch) {
RTC_CHECK_EQ(linear_aec_output_buf_[ch].size(), 160);
linear_aec_output_file_writer_->WriteSamples(
&linear_aec_output_buf_[ch][k], 1);
}
}
}
if (residual_echo_likelihood_graph_writer_.is_open()) {
auto stats = ap_->GetStatistics(true /*has_remote_tracks*/);
residual_echo_likelihood_graph_writer_
@ -342,6 +356,21 @@ void AudioProcessingSimulator::SetupOutput() {
settings_.processed_capture_samples);
}
if (settings_.linear_aec_output_filename) {
std::string filename;
if (settings_.store_intermediate_output) {
filename = GetIndexedOutputWavFilename(
*settings_.linear_aec_output_filename, output_reset_counter_);
} else {
filename = *settings_.linear_aec_output_filename;
}
linear_aec_output_file_writer_.reset(
new WavWriter(filename, 16000, out_config_.num_channels()));
linear_aec_output_buf_.resize(out_config_.num_channels());
}
if (settings_.reverse_output_filename) {
std::string filename;
if (settings_.store_intermediate_output) {
@ -410,6 +439,8 @@ void AudioProcessingSimulator::CreateAudioProcessor() {
apm_config.echo_canceller.mobile_mode = use_aecm;
apm_config.echo_canceller.use_legacy_aec = use_legacy_aec;
}
apm_config.echo_canceller.export_linear_aec_output =
!!settings_.linear_aec_output_filename;
RTC_CHECK(!(use_legacy_aec && settings_.aec_settings_filename))
<< "The legacy AEC cannot be configured using settings";
@ -421,9 +452,14 @@ void AudioProcessingSimulator::CreateAudioProcessor() {
std::cout << "Reading AEC Parameters from JSON input." << std::endl;
}
cfg = ReadAec3ConfigFromJsonFile(*settings_.aec_settings_filename);
echo_control_factory.reset(new EchoCanceller3Factory(cfg));
}
if (settings_.linear_aec_output_filename) {
cfg.filter.export_linear_aec_output = true;
}
echo_control_factory.reset(new EchoCanceller3Factory(cfg));
if (settings_.print_aec_parameter_values) {
if (!settings_.use_quiet_output) {
std::cout << "AEC settings:" << std::endl;

View file

@ -47,6 +47,7 @@ struct SimulationSettings {
absl::optional<std::string> input_filename;
absl::optional<std::string> reverse_input_filename;
absl::optional<std::string> artificial_nearend_filename;
absl::optional<std::string> linear_aec_output_filename;
absl::optional<bool> use_aec;
absl::optional<bool> use_aecm;
absl::optional<bool> use_ed; // Residual Echo Detector.
@ -156,6 +157,7 @@ class AudioProcessingSimulator {
std::unique_ptr<ChannelBuffer<float>> out_buf_;
std::unique_ptr<ChannelBuffer<float>> reverse_in_buf_;
std::unique_ptr<ChannelBuffer<float>> reverse_out_buf_;
std::vector<std::array<float, 160>> linear_aec_output_buf_;
StreamConfig in_config_;
StreamConfig out_config_;
StreamConfig reverse_in_config_;
@ -178,6 +180,7 @@ class AudioProcessingSimulator {
std::unique_ptr<ChannelBufferWavWriter> buffer_file_writer_;
std::unique_ptr<ChannelBufferWavWriter> reverse_buffer_file_writer_;
std::unique_ptr<ChannelBufferVectorWriter> buffer_memory_writer_;
std::unique_ptr<WavWriter> linear_aec_output_file_writer_;
ApiCallStatistics api_call_statistics_;
std::ofstream residual_echo_likelihood_graph_writer_;
int analog_mic_level_;

View file

@ -40,6 +40,7 @@ ABSL_FLAG(std::string,
artificial_nearend,
"",
"Artificial nearend wav filename");
ABSL_FLAG(std::string, linear_aec_output, "", "Linear AEC output wav filename");
ABSL_FLAG(int,
output_num_channels,
kParameterNotSpecifiedValue,
@ -364,6 +365,8 @@ SimulationSettings CreateSettings() {
&settings.reverse_output_filename);
SetSettingIfSpecified(absl::GetFlag(FLAGS_artificial_nearend),
&settings.artificial_nearend_filename);
SetSettingIfSpecified(absl::GetFlag(FLAGS_linear_aec_output),
&settings.linear_aec_output_filename);
SetSettingIfSpecified(absl::GetFlag(FLAGS_output_num_channels),
&settings.output_num_channels);
SetSettingIfSpecified(absl::GetFlag(FLAGS_reverse_output_num_channels),
@ -508,6 +511,19 @@ void PerformBasicParameterSanityChecks(const SimulationSettings& settings) {
"aec dump input string!\n");
}
ReportConditionalErrorAndExit(settings.use_aec && !(*settings.use_aec) &&
settings.linear_aec_output_filename,
"Error: The linear AEC ouput filename cannot "
"be specified without the AEC being active");
ReportConditionalErrorAndExit(
((settings.use_aec && *settings.use_aec && settings.use_legacy_aec &&
*settings.use_legacy_aec) ||
(settings.use_aecm && *settings.use_aecm)) &&
!!settings.linear_aec_output_filename,
"Error: The linear AEC ouput filename cannot be specified when the "
"legacy AEC or the AECm are used");
ReportConditionalErrorAndExit(
settings.use_aec && *settings.use_aec && settings.use_aecm &&
*settings.use_aecm,
@ -617,6 +633,11 @@ void PerformBasicParameterSanityChecks(const SimulationSettings& settings) {
!valid_wav_name(*settings.artificial_nearend_filename),
"Error: --artifical_nearend must be a valid .wav file name.\n");
ReportConditionalErrorAndExit(
settings.linear_aec_output_filename &&
(!valid_wav_name(*settings.linear_aec_output_filename)),
"Error: --linear_aec_output must be a valid .wav file name.\n");
ReportConditionalErrorAndExit(
WEBRTC_APM_DEBUG_DUMP == 0 && settings.dump_internal_data,
"Error: --dump_data cannot be set without proper build support.\n");

View file

@ -24,6 +24,10 @@ class MockEchoControl : public EchoControl {
MOCK_METHOD1(AnalyzeCapture, void(AudioBuffer* capture));
MOCK_METHOD2(ProcessCapture,
void(AudioBuffer* capture, bool echo_path_change));
MOCK_METHOD3(ProcessCapture,
void(AudioBuffer* capture,
AudioBuffer* linear_output,
bool echo_path_change));
MOCK_CONST_METHOD0(GetMetrics, EchoControl::Metrics());
MOCK_METHOD1(SetAudioBufferDelay, void(int delay_ms));
MOCK_CONST_METHOD0(ActiveProcessing, bool());