Render-side pre-processing in APM.

This CL adds a way to insert a custom render-side pre-processor to
APM. The pre-processor operates in full-band mode before anything
else. Currently the render processing chain is (if everything is
enabled):

Network --> [Pre processing] --> [Band split] -->
[IntelligibilityEnhancer] --> [Echo canceller (read-only)] -->
[Band merge] --> Playout

Since the render pre processor and capture post processor have the
same interface, I renamed webrtc::PostProcessing into
webrtc::CustomProcessing.

The old APM factory method PostProcessing will be deprecated and
dependencies updated as part of webrtc:8665

NOTRY=True

Bug: webrtc:8665
Change-Id: Ia381cbf12e336d6587406a14d77243d931f69a31
Reviewed-on: https://webrtc-review.googlesource.com/29201
Commit-Queue: Alex Loiko <aleloi@webrtc.org>
Reviewed-by: Per Åhgren <peah@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#21327}
This commit is contained in:
Alex Loiko
2017-12-18 16:02:40 +01:00
committed by Commit Bot
parent 88bc9d5e53
commit 5825aa673c
7 changed files with 112 additions and 33 deletions

View File

@ -172,8 +172,10 @@ webrtc::InternalAPMStreamsConfig ToStreamsConfig(
static_assert(AudioProcessing::kNoError == 0, "kNoError must be zero");
AudioProcessingImpl::ApmSubmoduleStates::ApmSubmoduleStates(
bool capture_post_processor_enabled)
: capture_post_processor_enabled_(capture_post_processor_enabled) {}
bool capture_post_processor_enabled,
bool render_pre_processor_enabled)
: capture_post_processor_enabled_(capture_post_processor_enabled),
render_pre_processor_enabled_(render_pre_processor_enabled) {}
bool AudioProcessingImpl::ApmSubmoduleStates::Update(
bool low_cut_filter_enabled,
@ -264,6 +266,11 @@ bool AudioProcessingImpl::ApmSubmoduleStates::RenderMultiBandSubModulesActive()
echo_controller_enabled_;
}
bool AudioProcessingImpl::ApmSubmoduleStates::RenderFullBandProcessingActive()
const {
return render_pre_processor_enabled_;
}
bool AudioProcessingImpl::ApmSubmoduleStates::RenderMultiBandProcessingActive()
const {
#if WEBRTC_INTELLIGIBILITY_ENHANCER
@ -294,9 +301,11 @@ struct AudioProcessingImpl::ApmPublicSubmodules {
struct AudioProcessingImpl::ApmPrivateSubmodules {
ApmPrivateSubmodules(NonlinearBeamformer* beamformer,
std::unique_ptr<PostProcessing> capture_post_processor)
std::unique_ptr<CustomProcessing> capture_post_processor,
std::unique_ptr<CustomProcessing> render_pre_processor)
: beamformer(beamformer),
capture_post_processor(std::move(capture_post_processor)) {}
capture_post_processor(std::move(capture_post_processor)),
render_pre_processor(std::move(render_pre_processor)) {}
// Accessed internally from capture or during initialization
std::unique_ptr<NonlinearBeamformer> beamformer;
std::unique_ptr<AgcManagerDirect> agc_manager;
@ -305,31 +314,43 @@ struct AudioProcessingImpl::ApmPrivateSubmodules {
std::unique_ptr<LevelController> level_controller;
std::unique_ptr<ResidualEchoDetector> residual_echo_detector;
std::unique_ptr<EchoControl> echo_controller;
std::unique_ptr<PostProcessing> capture_post_processor;
std::unique_ptr<CustomProcessing> capture_post_processor;
std::unique_ptr<CustomProcessing> render_pre_processor;
};
AudioProcessing* AudioProcessing::Create() {
webrtc::Config config;
return Create(config, nullptr, nullptr, nullptr);
return Create(config, nullptr, nullptr, nullptr, nullptr);
}
AudioProcessing* AudioProcessing::Create(const webrtc::Config& config) {
return Create(config, nullptr, nullptr, nullptr);
return Create(config, nullptr, nullptr, nullptr, nullptr);
}
AudioProcessing* AudioProcessing::Create(const webrtc::Config& config,
NonlinearBeamformer* beamformer) {
return Create(config, nullptr, nullptr, beamformer);
return Create(config, nullptr, nullptr, nullptr, beamformer);
}
AudioProcessing* AudioProcessing::Create(
const webrtc::Config& config,
std::unique_ptr<PostProcessing> capture_post_processor,
std::unique_ptr<CustomProcessing> capture_post_processor,
std::unique_ptr<EchoControlFactory> echo_control_factory,
NonlinearBeamformer* beamformer) {
return Create(config, std::move(capture_post_processor), nullptr,
std::move(echo_control_factory), beamformer);
}
AudioProcessing* AudioProcessing::Create(
const webrtc::Config& config,
std::unique_ptr<CustomProcessing> capture_post_processor,
std::unique_ptr<CustomProcessing> render_pre_processor,
std::unique_ptr<EchoControlFactory> echo_control_factory,
NonlinearBeamformer* beamformer) {
AudioProcessingImpl* apm = new rtc::RefCountedObject<AudioProcessingImpl>(
config, std::move(capture_post_processor),
std::move(echo_control_factory), beamformer);
std::move(render_pre_processor), std::move(echo_control_factory),
beamformer);
if (apm->Initialize() != kNoError) {
delete apm;
apm = nullptr;
@ -339,20 +360,22 @@ AudioProcessing* AudioProcessing::Create(
}
AudioProcessingImpl::AudioProcessingImpl(const webrtc::Config& config)
: AudioProcessingImpl(config, nullptr, nullptr, nullptr) {}
: AudioProcessingImpl(config, nullptr, nullptr, nullptr, nullptr) {}
AudioProcessingImpl::AudioProcessingImpl(
const webrtc::Config& config,
std::unique_ptr<PostProcessing> capture_post_processor,
std::unique_ptr<CustomProcessing> capture_post_processor,
std::unique_ptr<CustomProcessing> render_pre_processor,
std::unique_ptr<EchoControlFactory> echo_control_factory,
NonlinearBeamformer* beamformer)
: high_pass_filter_impl_(new HighPassFilterImpl(this)),
echo_control_factory_(std::move(echo_control_factory)),
submodule_states_(!!capture_post_processor),
submodule_states_(!!capture_post_processor, !!render_pre_processor),
public_submodules_(new ApmPublicSubmodules()),
private_submodules_(
new ApmPrivateSubmodules(beamformer,
std::move(capture_post_processor))),
std::move(capture_post_processor),
std::move(render_pre_processor))),
constants_(config.Get<ExperimentalAgc>().startup_min_volume,
config.Get<ExperimentalAgc>().clipped_level_min,
#if defined(WEBRTC_ANDROID) || defined(WEBRTC_IOS)
@ -405,6 +428,9 @@ AudioProcessingImpl::AudioProcessingImpl(
RTC_LOG(LS_INFO) << "Capture post processor activated: "
<< !!private_submodules_->capture_post_processor;
RTC_LOG(LS_INFO) << "Render pre processor activated: "
<< !!private_submodules_->render_pre_processor;
}
SetExtraOptions(config);
@ -560,6 +586,7 @@ int AudioProcessingImpl::InitializeLocked() {
InitializeEchoController();
InitializeGainController2();
InitializePostProcessor();
InitializePreProcessor();
if (aec_dump_) {
aec_dump_->WriteInitMessage(ToStreamsConfig(formats_.api_format));
@ -1345,7 +1372,8 @@ int AudioProcessingImpl::ProcessReverseStream(const float* const* src,
TRACE_EVENT0("webrtc", "AudioProcessing::ProcessReverseStream_StreamConfig");
rtc::CritScope cs(&crit_render_);
RETURN_ON_ERR(AnalyzeReverseStreamLocked(src, input_config, output_config));
if (submodule_states_.RenderMultiBandProcessingActive()) {
if (submodule_states_.RenderMultiBandProcessingActive() ||
submodule_states_.RenderFullBandProcessingActive()) {
render_.render_audio->CopyTo(formats_.api_format.reverse_output_stream(),
dest);
} else if (formats_.api_format.reverse_input_stream() !=
@ -1434,7 +1462,8 @@ int AudioProcessingImpl::ProcessReverseStream(AudioFrame* frame) {
render_.render_audio->DeinterleaveFrom(frame);
RETURN_ON_ERR(ProcessRenderStreamLocked());
render_.render_audio->InterleaveTo(
frame, submodule_states_.RenderMultiBandProcessingActive());
frame, submodule_states_.RenderMultiBandProcessingActive() ||
submodule_states_.RenderFullBandProcessingActive());
return kNoError;
}
@ -1443,6 +1472,10 @@ int AudioProcessingImpl::ProcessRenderStreamLocked() {
QueueNonbandedRenderAudio(render_buffer);
if (private_submodules_->render_pre_processor) {
private_submodules_->render_pre_processor->Process(render_buffer);
}
if (submodule_states_.RenderMultiBandSubModulesActive() &&
SampleRateSupportsMultiBand(
formats_.render_processing_format.sample_rate_hz())) {
@ -1792,6 +1825,14 @@ void AudioProcessingImpl::InitializePostProcessor() {
}
}
void AudioProcessingImpl::InitializePreProcessor() {
if (private_submodules_->render_pre_processor) {
private_submodules_->render_pre_processor->Initialize(
formats_.render_processing_format.sample_rate_hz(),
formats_.render_processing_format.num_channels());
}
}
void AudioProcessingImpl::MaybeUpdateHistograms() {
static const int kMinDiffDelayMs = 60;