AEC3: Adding a configurable render signal gain
Bug: webrtc:8671
Change-Id: I405d669517382ce195065caa3147eabace5ec18a
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/158669
Commit-Queue: Per Åhgren <peah@webrtc.org>
Reviewed-by: Gustaf Ullberg <gustaf@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#29647}
diff --git a/api/audio/echo_canceller3_config.h b/api/audio/echo_canceller3_config.h
index efd6bb5..3387526 100644
--- a/api/audio/echo_canceller3_config.h
+++ b/api/audio/echo_canceller3_config.h
@@ -113,6 +113,7 @@
float active_render_limit = 100.f;
float poor_excitation_render_limit = 150.f;
float poor_excitation_render_limit_ds8 = 20.f;
+ float render_power_gain_db = 0.f;
} render_levels;
struct EchoRemovalControl {
diff --git a/api/audio/echo_canceller3_config_json.cc b/api/audio/echo_canceller3_config_json.cc
index f7f115d..c6ee708 100644
--- a/api/audio/echo_canceller3_config_json.cc
+++ b/api/audio/echo_canceller3_config_json.cc
@@ -237,6 +237,8 @@
&cfg.render_levels.poor_excitation_render_limit);
ReadParam(section, "poor_excitation_render_limit_ds8",
&cfg.render_levels.poor_excitation_render_limit_ds8);
+ ReadParam(section, "render_power_gain_db",
+ &cfg.render_levels.render_power_gain_db);
}
if (rtc::GetValueFromJsonObject(aec3_root, "echo_removal_control",
@@ -459,7 +461,9 @@
ost << "\"poor_excitation_render_limit\": "
<< config.render_levels.poor_excitation_render_limit << ",";
ost << "\"poor_excitation_render_limit_ds8\": "
- << config.render_levels.poor_excitation_render_limit_ds8;
+ << config.render_levels.poor_excitation_render_limit_ds8 << ",";
+ ost << "\"render_power_gain_db\": "
+ << config.render_levels.render_power_gain_db;
ost << "},";
ost << "\"echo_removal_control\": {";
diff --git a/modules/audio_processing/aec3/render_delay_buffer.cc b/modules/audio_processing/aec3/render_delay_buffer.cc
index fb32035..091704c 100644
--- a/modules/audio_processing/aec3/render_delay_buffer.cc
+++ b/modules/audio_processing/aec3/render_delay_buffer.cc
@@ -13,6 +13,7 @@
#include <string.h>
#include <algorithm>
+#include <cmath>
#include <memory>
#include <numeric>
#include <vector>
@@ -70,6 +71,7 @@
std::unique_ptr<ApmDataDumper> data_dumper_;
const Aec3Optimization optimization_;
const EchoCanceller3Config config_;
+ const float render_linear_amplitude_gain_;
const rtc::LoggingSeverity delay_log_level_;
size_t down_sampling_factor_;
const int sub_block_size_;
@@ -118,6 +120,8 @@
new ApmDataDumper(rtc::AtomicOps::Increment(&instance_count_))),
optimization_(DetectOptimization()),
config_(config),
+ render_linear_amplitude_gain_(
+ std::pow(10.0f, config_.render_levels.render_power_gain_db / 20.f)),
delay_log_level_(config_.delay.log_warning_on_delay_changes
? rtc::LS_WARNING
: rtc::LS_INFO),
@@ -377,19 +381,37 @@
auto& ds = render_ds_;
auto& f = ffts_;
auto& s = spectra_;
+ const size_t num_bands = b.buffer[b.write].size();
+ const size_t num_render_channels = b.buffer[b.write][0].size();
RTC_DCHECK_EQ(block.size(), b.buffer[b.write].size());
- for (size_t k = 0; k < block.size(); ++k) {
- RTC_DCHECK_EQ(block[k].size(), b.buffer[b.write][k].size());
- std::copy(block[k].begin(), block[k].end(), b.buffer[b.write][k].begin());
+ for (size_t band = 0; band < num_bands; ++band) {
+ RTC_DCHECK_EQ(block[band].size(), num_render_channels);
+ RTC_DCHECK_EQ(b.buffer[b.write][band].size(), num_render_channels);
+ for (size_t ch = 0; ch < num_render_channels; ++ch) {
+ RTC_DCHECK_EQ(block[band][ch].size(), b.buffer[b.write][band][ch].size());
+ std::copy(block[band][ch].begin(), block[band][ch].end(),
+ b.buffer[b.write][band][ch].begin());
+ }
}
- render_decimator_.Decimate(block[0],
+ if (render_linear_amplitude_gain_ != 1.f) {
+ for (size_t band = 0; band < num_bands; ++band) {
+ for (size_t ch = 0; ch < num_render_channels; ++ch) {
+ for (size_t k = 0; k < 64; ++k) {
+ b.buffer[b.write][band][ch][k] *= render_linear_amplitude_gain_;
+ }
+ }
+ }
+ }
+
+ render_decimator_.Decimate(b.buffer[b.write][0],
config_.delay.downmix_before_delay_estimation, ds);
data_dumper_->DumpWav("aec3_render_decimator_output", ds.size(), ds.data(),
16000 / down_sampling_factor_, 1);
std::copy(ds.rbegin(), ds.rend(), lr.buffer.begin() + lr.write);
- for (size_t channel = 0; channel < block[0].size(); ++channel) {
- fft_.PaddedFft(block[0][channel], b.buffer[previous_write][0][channel],
+ for (size_t channel = 0; channel < b.buffer[b.write][0].size(); ++channel) {
+ fft_.PaddedFft(b.buffer[b.write][0][channel],
+ b.buffer[previous_write][0][channel],
&f.buffer[f.write][channel]);
f.buffer[f.write][channel].Spectrum(optimization_,
s.buffer[s.write][channel]);
