Make NoiseSuppression not a processing component (bit exact).
BUG=webrtc:5298
patch from issue 1490333004 at patchset 1 (http://crrev.com/1490333004#ps1)
Review URL: https://codereview.webrtc.org/1507683006
Cr-Commit-Position: refs/heads/master@{#10944}
diff --git a/webrtc/modules/audio_processing/audio_processing_impl.cc b/webrtc/modules/audio_processing/audio_processing_impl.cc
index a38d73e..805e05e 100644
--- a/webrtc/modules/audio_processing/audio_processing_impl.cc
+++ b/webrtc/modules/audio_processing/audio_processing_impl.cc
@@ -76,38 +76,6 @@
}
} // namespace
-struct AudioProcessingImpl::ApmPublicSubmodules {
- ApmPublicSubmodules()
- : echo_cancellation(nullptr),
- echo_control_mobile(nullptr),
- gain_control(nullptr),
- level_estimator(nullptr),
- noise_suppression(nullptr),
- voice_detection(nullptr) {}
- // Accessed externally of APM without any lock acquired.
- EchoCancellationImpl* echo_cancellation;
- EchoControlMobileImpl* echo_control_mobile;
- GainControlImpl* gain_control;
- rtc::scoped_ptr<HighPassFilterImpl> high_pass_filter;
- LevelEstimatorImpl* level_estimator;
- NoiseSuppressionImpl* noise_suppression;
- VoiceDetectionImpl* voice_detection;
- rtc::scoped_ptr<GainControlForNewAgc> gain_control_for_new_agc;
-
- // Accessed internally from both render and capture.
- rtc::scoped_ptr<TransientSuppressor> transient_suppressor;
- rtc::scoped_ptr<IntelligibilityEnhancer> intelligibility_enhancer;
-};
-
-struct AudioProcessingImpl::ApmPrivateSubmodules {
- explicit ApmPrivateSubmodules(Beamformer<float>* beamformer)
- : beamformer(beamformer) {}
- // Accessed internally from capture or during initialization
- std::list<ProcessingComponent*> component_list;
- rtc::scoped_ptr<Beamformer<float>> beamformer;
- rtc::scoped_ptr<AgcManagerDirect> agc_manager;
-};
-
// Throughout webrtc, it's assumed that success is represented by zero.
static_assert(AudioProcessing::kNoError == 0, "kNoError must be zero");
@@ -177,6 +145,37 @@
int volume_;
};
+struct AudioProcessingImpl::ApmPublicSubmodules {
+ ApmPublicSubmodules()
+ : echo_cancellation(nullptr),
+ echo_control_mobile(nullptr),
+ gain_control(nullptr),
+ level_estimator(nullptr),
+ voice_detection(nullptr) {}
+ // Accessed externally of APM without any lock acquired.
+ EchoCancellationImpl* echo_cancellation;
+ EchoControlMobileImpl* echo_control_mobile;
+ GainControlImpl* gain_control;
+ rtc::scoped_ptr<HighPassFilterImpl> high_pass_filter;
+ LevelEstimatorImpl* level_estimator;
+ rtc::scoped_ptr<NoiseSuppressionImpl> noise_suppression;
+ VoiceDetectionImpl* voice_detection;
+ rtc::scoped_ptr<GainControlForNewAgc> gain_control_for_new_agc;
+
+ // Accessed internally from both render and capture.
+ rtc::scoped_ptr<TransientSuppressor> transient_suppressor;
+ rtc::scoped_ptr<IntelligibilityEnhancer> intelligibility_enhancer;
+};
+
+struct AudioProcessingImpl::ApmPrivateSubmodules {
+ explicit ApmPrivateSubmodules(Beamformer<float>* beamformer)
+ : beamformer(beamformer) {}
+ // Accessed internally from capture or during initialization
+ std::list<ProcessingComponent*> component_list;
+ rtc::scoped_ptr<Beamformer<float>> beamformer;
+ rtc::scoped_ptr<AgcManagerDirect> agc_manager;
+};
+
const int AudioProcessing::kNativeSampleRatesHz[] = {
AudioProcessing::kSampleRate8kHz,
AudioProcessing::kSampleRate16kHz,
@@ -246,8 +245,8 @@
new HighPassFilterImpl(&crit_capture_));
public_submodules_->level_estimator =
new LevelEstimatorImpl(this, &crit_capture_);
- public_submodules_->noise_suppression =
- new NoiseSuppressionImpl(this, &crit_capture_);
+ public_submodules_->noise_suppression.reset(
+ new NoiseSuppressionImpl(&crit_capture_));
public_submodules_->voice_detection =
new VoiceDetectionImpl(this, &crit_capture_);
public_submodules_->gain_control_for_new_agc.reset(
@@ -262,8 +261,6 @@
private_submodules_->component_list.push_back(
public_submodules_->level_estimator);
private_submodules_->component_list.push_back(
- public_submodules_->noise_suppression);
- private_submodules_->component_list.push_back(
public_submodules_->voice_detection);
}
@@ -396,14 +393,11 @@
}
InitializeExperimentalAgc();
-
InitializeTransient();
-
InitializeBeamformer();
-
InitializeIntelligibility();
-
InitializeHighPassFilter();
+ InitializeNoiseSuppression();
#ifdef WEBRTC_AUDIOPROC_DEBUG_DUMP
if (debug_dump_.debug_file->Open()) {
@@ -768,14 +762,14 @@
public_submodules_->high_pass_filter->ProcessCaptureAudio(ca);
RETURN_ON_ERR(public_submodules_->gain_control->AnalyzeCaptureAudio(ca));
- RETURN_ON_ERR(public_submodules_->noise_suppression->AnalyzeCaptureAudio(ca));
+ public_submodules_->noise_suppression->AnalyzeCaptureAudio(ca);
RETURN_ON_ERR(public_submodules_->echo_cancellation->ProcessCaptureAudio(ca));
if (public_submodules_->echo_control_mobile->is_enabled() &&
public_submodules_->noise_suppression->is_enabled()) {
ca->CopyLowPassToReference();
}
- RETURN_ON_ERR(public_submodules_->noise_suppression->ProcessCaptureAudio(ca));
+ public_submodules_->noise_suppression->ProcessCaptureAudio(ca);
RETURN_ON_ERR(
public_submodules_->echo_control_mobile->ProcessCaptureAudio(ca));
RETURN_ON_ERR(public_submodules_->voice_detection->ProcessCaptureAudio(ca));
@@ -1158,7 +1152,7 @@
NoiseSuppression* AudioProcessingImpl::noise_suppression() const {
// Adding a lock here has no effect as it allows any access to the submodule
// from the returned pointer.
- return public_submodules_->noise_suppression;
+ return public_submodules_->noise_suppression.get();
}
VoiceDetection* AudioProcessingImpl::voice_detection() const {
@@ -1181,6 +1175,9 @@
if (public_submodules_->high_pass_filter->is_enabled()) {
enabled_count++;
}
+ if (public_submodules_->noise_suppression->is_enabled()) {
+ enabled_count++;
+ }
// Data is unchanged if no components are enabled, or if only
// public_submodules_->level_estimator
@@ -1300,6 +1297,11 @@
proc_sample_rate_hz());
}
+void AudioProcessingImpl::InitializeNoiseSuppression() {
+ public_submodules_->noise_suppression->Initialize(num_output_channels(),
+ proc_sample_rate_hz());
+}
+
void AudioProcessingImpl::MaybeUpdateHistograms() {
static const int kMinDiffDelayMs = 60;
diff --git a/webrtc/modules/audio_processing/audio_processing_impl.h b/webrtc/modules/audio_processing/audio_processing_impl.h
index 1ca5a8c..c4d5f08 100644
--- a/webrtc/modules/audio_processing/audio_processing_impl.h
+++ b/webrtc/modules/audio_processing/audio_processing_impl.h
@@ -39,18 +39,6 @@
template<typename T>
class Beamformer;
-class EchoCancellationImpl;
-class EchoControlMobileImpl;
-class GainControlImpl;
-class GainControlForNewAgc;
-class HighPassFilterImpl;
-class LevelEstimatorImpl;
-class NoiseSuppressionImpl;
-class ProcessingComponent;
-class TransientSuppressor;
-class VoiceDetectionImpl;
-class IntelligibilityEnhancer;
-
class AudioProcessingImpl : public AudioProcessing {
public:
// Methods forcing APM to run in a single-threaded manner.
@@ -195,6 +183,8 @@
EXCLUSIVE_LOCKS_REQUIRED(crit_render_, crit_capture_);
void InitializeHighPassFilter()
EXCLUSIVE_LOCKS_REQUIRED(crit_capture_);
+ void InitializeNoiseSuppression()
+ EXCLUSIVE_LOCKS_REQUIRED(crit_capture_);
int InitializeLocked(const ProcessingConfig& config)
EXCLUSIVE_LOCKS_REQUIRED(crit_render_, crit_capture_);
diff --git a/webrtc/modules/audio_processing/high_pass_filter_impl.cc b/webrtc/modules/audio_processing/high_pass_filter_impl.cc
index 795dcbd..0db794f 100644
--- a/webrtc/modules/audio_processing/high_pass_filter_impl.cc
+++ b/webrtc/modules/audio_processing/high_pass_filter_impl.cc
@@ -89,6 +89,7 @@
HighPassFilterImpl::~HighPassFilterImpl() {}
void HighPassFilterImpl::Initialize(int channels, int sample_rate_hz) {
+ RTC_DCHECK_LE(0, channels);
std::vector<rtc::scoped_ptr<BiquadFilter>> new_filters(channels);
for (int i = 0; i < channels; i++) {
new_filters[i].reset(new BiquadFilter(sample_rate_hz));
@@ -98,6 +99,7 @@
}
void HighPassFilterImpl::ProcessCaptureAudio(AudioBuffer* audio) {
+ RTC_DCHECK(audio);
rtc::CritScope cs(crit_);
if (!enabled_) {
return;
diff --git a/webrtc/modules/audio_processing/high_pass_filter_impl.h b/webrtc/modules/audio_processing/high_pass_filter_impl.h
index b2a0717..d6e84fd 100644
--- a/webrtc/modules/audio_processing/high_pass_filter_impl.h
+++ b/webrtc/modules/audio_processing/high_pass_filter_impl.h
@@ -11,6 +11,7 @@
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_HIGH_PASS_FILTER_IMPL_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_HIGH_PASS_FILTER_IMPL_H_
+#include "webrtc/base/constructormagic.h"
#include "webrtc/base/criticalsection.h"
#include "webrtc/base/scoped_ptr.h"
#include "webrtc/modules/audio_processing/include/audio_processing.h"
@@ -37,6 +38,7 @@
rtc::CriticalSection* const crit_ = nullptr;
bool enabled_ GUARDED_BY(crit_) = false;
std::vector<rtc::scoped_ptr<BiquadFilter>> filters_ GUARDED_BY(crit_);
+ RTC_DISALLOW_IMPLICIT_CONSTRUCTORS(HighPassFilterImpl);
};
} // namespace webrtc
diff --git a/webrtc/modules/audio_processing/noise_suppression_impl.cc b/webrtc/modules/audio_processing/noise_suppression_impl.cc
index 837585f..d4f8bbb 100644
--- a/webrtc/modules/audio_processing/noise_suppression_impl.cc
+++ b/webrtc/modules/audio_processing/noise_suppression_impl.cc
@@ -10,84 +10,96 @@
#include "webrtc/modules/audio_processing/noise_suppression_impl.h"
-#include <assert.h>
-
#include "webrtc/modules/audio_processing/audio_buffer.h"
#if defined(WEBRTC_NS_FLOAT)
#include "webrtc/modules/audio_processing/ns/noise_suppression.h"
+#define NS_CREATE WebRtcNs_Create
+#define NS_FREE WebRtcNs_Free
+#define NS_INIT WebRtcNs_Init
+#define NS_SET_POLICY WebRtcNs_set_policy
+typedef NsHandle NsState;
#elif defined(WEBRTC_NS_FIXED)
#include "webrtc/modules/audio_processing/ns/noise_suppression_x.h"
+#define NS_CREATE WebRtcNsx_Create
+#define NS_FREE WebRtcNsx_Free
+#define NS_INIT WebRtcNsx_Init
+#define NS_SET_POLICY WebRtcNsx_set_policy
+typedef NsxHandle NsState;
#endif
-
namespace webrtc {
-
-#if defined(WEBRTC_NS_FLOAT)
-typedef NsHandle Handle;
-#elif defined(WEBRTC_NS_FIXED)
-typedef NsxHandle Handle;
-#endif
-
-namespace {
-int MapSetting(NoiseSuppression::Level level) {
- switch (level) {
- case NoiseSuppression::kLow:
- return 0;
- case NoiseSuppression::kModerate:
- return 1;
- case NoiseSuppression::kHigh:
- return 2;
- case NoiseSuppression::kVeryHigh:
- return 3;
+class NoiseSuppressionImpl::Suppressor {
+ public:
+ explicit Suppressor(int sample_rate_hz) {
+ state_ = NS_CREATE();
+ RTC_CHECK(state_);
+ int error = NS_INIT(state_, sample_rate_hz);
+ RTC_DCHECK_EQ(0, error);
}
- assert(false);
- return -1;
-}
-} // namespace
+ ~Suppressor() {
+ NS_FREE(state_);
+ }
+ NsState* state() { return state_; }
+ private:
+ NsState* state_ = nullptr;
+ RTC_DISALLOW_IMPLICIT_CONSTRUCTORS(Suppressor);
+};
-NoiseSuppressionImpl::NoiseSuppressionImpl(const AudioProcessing* apm,
- rtc::CriticalSection* crit)
- : ProcessingComponent(), apm_(apm), crit_(crit), level_(kModerate) {
- RTC_DCHECK(apm);
+NoiseSuppressionImpl::NoiseSuppressionImpl(rtc::CriticalSection* crit)
+ : crit_(crit) {
RTC_DCHECK(crit);
}
NoiseSuppressionImpl::~NoiseSuppressionImpl() {}
+void NoiseSuppressionImpl::Initialize(int channels, int sample_rate_hz) {
+ RTC_DCHECK_LE(0, channels);
+ std::vector<rtc::scoped_ptr<Suppressor>> new_suppressors(channels);
+ for (int i = 0; i < channels; i++) {
+ new_suppressors[i].reset(new Suppressor(sample_rate_hz));
+ }
+ rtc::CritScope cs(crit_);
+ suppressors_.swap(new_suppressors);
+ set_level(level_);
+}
+
int NoiseSuppressionImpl::AnalyzeCaptureAudio(AudioBuffer* audio) {
+ RTC_DCHECK(audio);
#if defined(WEBRTC_NS_FLOAT)
- if (!is_component_enabled()) {
+ rtc::CritScope cs(crit_);
+ if (!enabled_) {
return AudioProcessing::kNoError;
}
- assert(audio->num_frames_per_band() <= 160);
- assert(audio->num_channels() == num_handles());
- for (int i = 0; i < num_handles(); ++i) {
- Handle* my_handle = static_cast<Handle*>(handle(i));
-
- WebRtcNs_Analyze(my_handle, audio->split_bands_const_f(i)[kBand0To8kHz]);
+ RTC_DCHECK_GE(160u, audio->num_frames_per_band());
+ RTC_DCHECK_EQ(suppressors_.size(),
+ static_cast<size_t>(audio->num_channels()));
+ for (size_t i = 0; i < suppressors_.size(); i++) {
+ WebRtcNs_Analyze(suppressors_[i]->state(),
+ audio->split_bands_const_f(i)[kBand0To8kHz]);
}
#endif
return AudioProcessing::kNoError;
}
int NoiseSuppressionImpl::ProcessCaptureAudio(AudioBuffer* audio) {
+ RTC_DCHECK(audio);
rtc::CritScope cs(crit_);
- if (!is_component_enabled()) {
+ if (!enabled_) {
return AudioProcessing::kNoError;
}
- assert(audio->num_frames_per_band() <= 160);
- assert(audio->num_channels() == num_handles());
- for (int i = 0; i < num_handles(); ++i) {
- Handle* my_handle = static_cast<Handle*>(handle(i));
+ RTC_DCHECK_GE(160u, audio->num_frames_per_band());
+ RTC_DCHECK_EQ(suppressors_.size(),
+ static_cast<size_t>(audio->num_channels()));
+ for (size_t i = 0; i < suppressors_.size(); i++) {
#if defined(WEBRTC_NS_FLOAT)
- WebRtcNs_Process(my_handle,
+ WebRtcNs_Process(suppressors_[i]->state(),
audio->split_bands_const_f(i),
audio->num_bands(),
audio->split_bands_f(i));
#elif defined(WEBRTC_NS_FIXED)
- WebRtcNsx_Process(my_handle,
+ WebRtcNsx_Process(suppressors_[i]->state(),
audio->split_bands_const(i),
audio->num_bands(),
audio->split_bands(i));
@@ -98,22 +110,40 @@
int NoiseSuppressionImpl::Enable(bool enable) {
rtc::CritScope cs(crit_);
- return EnableComponent(enable);
+ enabled_ = enable;
+ return AudioProcessing::kNoError;
}
bool NoiseSuppressionImpl::is_enabled() const {
rtc::CritScope cs(crit_);
- return is_component_enabled();
+ return enabled_;
}
int NoiseSuppressionImpl::set_level(Level level) {
rtc::CritScope cs(crit_);
- if (MapSetting(level) == -1) {
- return AudioProcessing::kBadParameterError;
+ int policy = 1;
+ switch (level) {
+ case NoiseSuppression::kLow:
+ policy = 0;
+ break;
+ case NoiseSuppression::kModerate:
+ policy = 1;
+ break;
+ case NoiseSuppression::kHigh:
+ policy = 2;
+ break;
+ case NoiseSuppression::kVeryHigh:
+ policy = 3;
+ break;
+ default:
+ RTC_NOTREACHED();
}
-
level_ = level;
- return Configure();
+ for (auto& suppressor : suppressors_) {
+ int error = NS_SET_POLICY(suppressor->state(), policy);
+ RTC_DCHECK_EQ(0, error);
+ }
+ return AudioProcessing::kNoError;
}
NoiseSuppression::Level NoiseSuppressionImpl::level() const {
@@ -125,61 +155,18 @@
rtc::CritScope cs(crit_);
#if defined(WEBRTC_NS_FLOAT)
float probability_average = 0.0f;
- for (int i = 0; i < num_handles(); i++) {
- Handle* my_handle = static_cast<Handle*>(handle(i));
- probability_average += WebRtcNs_prior_speech_probability(my_handle);
+ for (auto& suppressor : suppressors_) {
+ probability_average +=
+ WebRtcNs_prior_speech_probability(suppressor->state());
}
- return probability_average / num_handles();
+ if (suppressors_.size() > 0) {
+ probability_average /= suppressors_.size();
+ }
+ return probability_average;
#elif defined(WEBRTC_NS_FIXED)
+ // TODO(peah): Returning error code as a float! Remove this.
// Currently not available for the fixed point implementation.
return AudioProcessing::kUnsupportedFunctionError;
#endif
}
-
-void* NoiseSuppressionImpl::CreateHandle() const {
-#if defined(WEBRTC_NS_FLOAT)
- return WebRtcNs_Create();
-#elif defined(WEBRTC_NS_FIXED)
- return WebRtcNsx_Create();
-#endif
-}
-
-void NoiseSuppressionImpl::DestroyHandle(void* handle) const {
-#if defined(WEBRTC_NS_FLOAT)
- WebRtcNs_Free(static_cast<Handle*>(handle));
-#elif defined(WEBRTC_NS_FIXED)
- WebRtcNsx_Free(static_cast<Handle*>(handle));
-#endif
-}
-
-int NoiseSuppressionImpl::InitializeHandle(void* handle) const {
-#if defined(WEBRTC_NS_FLOAT)
- return WebRtcNs_Init(static_cast<Handle*>(handle),
- apm_->proc_sample_rate_hz());
-#elif defined(WEBRTC_NS_FIXED)
- return WebRtcNsx_Init(static_cast<Handle*>(handle),
- apm_->proc_sample_rate_hz());
-#endif
-}
-
-int NoiseSuppressionImpl::ConfigureHandle(void* handle) const {
- rtc::CritScope cs(crit_);
-#if defined(WEBRTC_NS_FLOAT)
- return WebRtcNs_set_policy(static_cast<Handle*>(handle),
- MapSetting(level_));
-#elif defined(WEBRTC_NS_FIXED)
- return WebRtcNsx_set_policy(static_cast<Handle*>(handle),
- MapSetting(level_));
-#endif
-}
-
-int NoiseSuppressionImpl::num_handles_required() const {
- return apm_->num_output_channels();
-}
-
-int NoiseSuppressionImpl::GetHandleError(void* handle) const {
- // The NS has no get_error() function.
- assert(handle != NULL);
- return AudioProcessing::kUnspecifiedError;
-}
} // namespace webrtc
diff --git a/webrtc/modules/audio_processing/noise_suppression_impl.h b/webrtc/modules/audio_processing/noise_suppression_impl.h
index 1564fe5..a3b3460 100644
--- a/webrtc/modules/audio_processing/noise_suppression_impl.h
+++ b/webrtc/modules/audio_processing/noise_suppression_impl.h
@@ -11,49 +11,40 @@
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_NOISE_SUPPRESSION_IMPL_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_NOISE_SUPPRESSION_IMPL_H_
+#include "webrtc/base/constructormagic.h"
#include "webrtc/base/criticalsection.h"
+#include "webrtc/base/scoped_ptr.h"
#include "webrtc/modules/audio_processing/include/audio_processing.h"
-#include "webrtc/modules/audio_processing/processing_component.h"
namespace webrtc {
class AudioBuffer;
-class NoiseSuppressionImpl : public NoiseSuppression,
- public ProcessingComponent {
+class NoiseSuppressionImpl : public NoiseSuppression {
public:
- NoiseSuppressionImpl(const AudioProcessing* apm, rtc::CriticalSection* crit);
- virtual ~NoiseSuppressionImpl();
+ explicit NoiseSuppressionImpl(rtc::CriticalSection* crit);
+ ~NoiseSuppressionImpl() override;
+ // TODO(peah): Fold into ctor, once public API is removed.
+ void Initialize(int channels, int sample_rate_hz);
int AnalyzeCaptureAudio(AudioBuffer* audio);
int ProcessCaptureAudio(AudioBuffer* audio);
// NoiseSuppression implementation.
+ int Enable(bool enable) override;
bool is_enabled() const override;
- float speech_probability() const override;
+ int set_level(Level level) override;
Level level() const override;
+ float speech_probability() const override;
private:
- // NoiseSuppression implementation.
- int Enable(bool enable) override;
- int set_level(Level level) override;
-
- // ProcessingComponent implementation.
- void* CreateHandle() const override;
- int InitializeHandle(void* handle) const override;
- int ConfigureHandle(void* handle) const override;
- void DestroyHandle(void* handle) const override;
- int num_handles_required() const override;
- int GetHandleError(void* handle) const override;
-
- // Not guarded as its public API is thread safe.
- const AudioProcessing* apm_;
-
+ class Suppressor;
rtc::CriticalSection* const crit_;
-
- Level level_ GUARDED_BY(crit_);
+ bool enabled_ GUARDED_BY(crit_) = false;
+ Level level_ GUARDED_BY(crit_) = kModerate;
+ std::vector<rtc::scoped_ptr<Suppressor>> suppressors_ GUARDED_BY(crit_);
+ RTC_DISALLOW_IMPLICIT_CONSTRUCTORS(NoiseSuppressionImpl);
};
-
} // namespace webrtc
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_NOISE_SUPPRESSION_IMPL_H_
