Only adapt AGC when the desired signal is present
Take the 50% quantile of the mask and compare it to certain threshold to determine if the desired signal is present. A hold is applied to avoid fast switching between states.
is_signal_present_ has been plotted and looks as expected. The AGC adaptation sounds promising, specially for the cases when the speaker fades in and out from the beam direction.
R=andrew@webrtc.org
Review URL: https://webrtc-codereview.appspot.com/28329005
git-svn-id: http://webrtc.googlecode.com/svn/trunk@8078 4adac7df-926f-26a2-2b94-8c16560cd09d
diff --git a/webrtc/modules/audio_processing/audio_processing_impl.cc b/webrtc/modules/audio_processing/audio_processing_impl.cc
index 5d7de3a..2fa6f28 100644
--- a/webrtc/modules/audio_processing/audio_processing_impl.cc
+++ b/webrtc/modules/audio_processing/audio_processing_impl.cc
@@ -137,11 +137,16 @@
AudioProcessing* AudioProcessing::Create() {
Config config;
- return Create(config);
+ return Create(config, nullptr);
}
AudioProcessing* AudioProcessing::Create(const Config& config) {
- AudioProcessingImpl* apm = new AudioProcessingImpl(config);
+ return Create(config, nullptr);
+}
+
+AudioProcessing* AudioProcessing::Create(const Config& config,
+ Beamformer* beamformer) {
+ AudioProcessingImpl* apm = new AudioProcessingImpl(config, beamformer);
if (apm->Initialize() != kNoError) {
delete apm;
apm = NULL;
@@ -151,6 +156,10 @@
}
AudioProcessingImpl::AudioProcessingImpl(const Config& config)
+ : AudioProcessingImpl(config, nullptr) {}
+
+AudioProcessingImpl::AudioProcessingImpl(const Config& config,
+ Beamformer* beamformer)
: echo_cancellation_(NULL),
echo_control_mobile_(NULL),
gain_control_(NULL),
@@ -181,6 +190,7 @@
#endif
transient_suppressor_enabled_(config.Get<ExperimentalNs>().enabled),
beamformer_enabled_(config.Get<Beamforming>().enabled),
+ beamformer_(beamformer),
array_geometry_(config.Get<Beamforming>().array_geometry) {
echo_cancellation_ = new EchoCancellationImpl(this, crit_);
component_list_.push_back(echo_cancellation_);
@@ -330,6 +340,11 @@
num_reverse_channels > 2 || num_reverse_channels < 1) {
return kBadNumberChannelsError;
}
+ if (beamformer_enabled_ &&
+ (static_cast<size_t>(num_input_channels) != array_geometry_.size() ||
+ num_output_channels > 1)) {
+ return kBadNumberChannelsError;
+ }
fwd_in_format_.set(input_sample_rate_hz, num_input_channels);
fwd_out_format_.set(output_sample_rate_hz, num_output_channels);
@@ -395,11 +410,6 @@
num_reverse_channels == rev_in_format_.num_channels()) {
return kNoError;
}
- if (beamformer_enabled_ &&
- (static_cast<size_t>(num_input_channels) != array_geometry_.size() ||
- num_output_channels > 1)) {
- return kBadNumberChannelsError;
- }
return InitializeLocked(input_sample_rate_hz,
output_sample_rate_hz,
reverse_sample_rate_hz,
@@ -622,7 +632,9 @@
RETURN_ON_ERR(echo_control_mobile_->ProcessCaptureAudio(ca));
RETURN_ON_ERR(voice_detection_->ProcessCaptureAudio(ca));
- if (use_new_agc_ && gain_control_->is_enabled()) {
+ if (use_new_agc_ &&
+ gain_control_->is_enabled() &&
+ (!beamformer_enabled_ || beamformer_->is_target_present())) {
agc_manager_->Process(ca->split_bands_const(0)[kBand0To8kHz],
ca->samples_per_split_channel(),
split_rate_);
@@ -990,9 +1002,10 @@
void AudioProcessingImpl::InitializeBeamformer() {
if (beamformer_enabled_) {
#ifdef WEBRTC_BEAMFORMER
- beamformer_.reset(new Beamformer(kChunkSizeMs,
- split_rate_,
- array_geometry_));
+ if (!beamformer_) {
+ beamformer_.reset(new Beamformer(array_geometry_));
+ }
+ beamformer_->Initialize(kChunkSizeMs, split_rate_);
#else
assert(false);
#endif