blob: 4f2d9985475d7e0d7bf927f245b856546688b21d [file] [log] [blame]
/*
* Copyright (c) 2016 The WebRTC project authors. All Rights Reserved.
*
* Use of this source code is governed by a BSD-style license
* that can be found in the LICENSE file in the root of the source
* tree. An additional intellectual property rights grant can be found
* in the file PATENTS. All contributing project authors may
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/modules/audio_processing/level_controller/signal_classifier.h"
#include <algorithm>
#include <numeric>
#include <vector>
#include "webrtc/base/array_view.h"
#include "webrtc/base/constructormagic.h"
#include "webrtc/modules/audio_processing/aec/aec_rdft.h"
#include "webrtc/modules/audio_processing/audio_buffer.h"
#include "webrtc/modules/audio_processing/level_controller/down_sampler.h"
#include "webrtc/modules/audio_processing/level_controller/noise_spectrum_estimator.h"
#include "webrtc/modules/audio_processing/logging/apm_data_dumper.h"
namespace webrtc {
namespace {
void RemoveDcLevel(rtc::ArrayView<float> x) {
RTC_DCHECK_LT(0u, x.size());
float mean = std::accumulate(x.data(), x.data() + x.size(), 0.f);
mean /= x.size();
for (float& v : x) {
v -= mean;
}
}
void PowerSpectrum(rtc::ArrayView<const float> x,
rtc::ArrayView<float> spectrum) {
RTC_DCHECK_EQ(65u, spectrum.size());
RTC_DCHECK_EQ(128u, x.size());
float X[128];
std::copy(x.data(), x.data() + x.size(), X);
aec_rdft_forward_128(X);
float* X_p = X;
RTC_DCHECK_EQ(X_p, &X[0]);
spectrum[0] = (*X_p) * (*X_p);
++X_p;
RTC_DCHECK_EQ(X_p, &X[1]);
spectrum[64] = (*X_p) * (*X_p);
for (int k = 1; k < 64; ++k) {
++X_p;
RTC_DCHECK_EQ(X_p, &X[2 * k]);
spectrum[k] = (*X_p) * (*X_p);
++X_p;
RTC_DCHECK_EQ(X_p, &X[2 * k + 1]);
spectrum[k] += (*X_p) * (*X_p);
}
}
webrtc::SignalClassifier::SignalType ClassifySignal(
rtc::ArrayView<const float> signal_spectrum,
rtc::ArrayView<const float> noise_spectrum,
ApmDataDumper* data_dumper) {
int num_stationary_bands = 0;
int num_highly_nonstationary_bands = 0;
// Detect stationary and highly nonstationary bands.
for (size_t k = 1; k < 40; k++) {
if (signal_spectrum[k] < 3 * noise_spectrum[k] &&
signal_spectrum[k] * 3 > noise_spectrum[k]) {
++num_stationary_bands;
} else if (signal_spectrum[k] > 9 * noise_spectrum[k]) {
++num_highly_nonstationary_bands;
}
}
data_dumper->DumpRaw("lc_num_stationary_bands", 1, &num_stationary_bands);
data_dumper->DumpRaw("lc_num_highly_nonstationary_bands", 1,
&num_highly_nonstationary_bands);
// Use the detected number of bands to classify the overall signal
// stationarity.
if (num_stationary_bands > 15) {
return SignalClassifier::SignalType::kStationary;
} else if (num_highly_nonstationary_bands > 15) {
return SignalClassifier::SignalType::kHighlyNonStationary;
} else {
return SignalClassifier::SignalType::kNonStationary;
}
}
} // namespace
void SignalClassifier::FrameExtender::ExtendFrame(
rtc::ArrayView<const float> x,
rtc::ArrayView<float> x_extended) {
RTC_DCHECK_EQ(x_old_.size() + x.size(), x_extended.size());
std::copy(x_old_.data(), x_old_.data() + x_old_.size(), x_extended.data());
std::copy(x.data(), x.data() + x.size(), x_extended.data() + x_old_.size());
std::copy(x_extended.data() + x_extended.size() - x_old_.size(),
x_extended.data() + x_extended.size(), x_old_.data());
}
SignalClassifier::SignalClassifier(ApmDataDumper* data_dumper)
: data_dumper_(data_dumper),
down_sampler_(data_dumper_),
noise_spectrum_estimator_(data_dumper_) {
Initialize(AudioProcessing::kSampleRate48kHz);
}
SignalClassifier::~SignalClassifier() {}
void SignalClassifier::Initialize(int sample_rate_hz) {
aec_rdft_init();
down_sampler_.Initialize(sample_rate_hz);
noise_spectrum_estimator_.Initialize();
frame_extender_.reset(new FrameExtender(80, 128));
sample_rate_hz_ = sample_rate_hz;
initialization_frames_left_ = 2;
consistent_classification_counter_ = 3;
last_signal_type_ = SignalClassifier::SignalType::kNonStationary;
}
void SignalClassifier::Analyze(const AudioBuffer& audio,
SignalType* signal_type) {
RTC_DCHECK_EQ(audio.num_frames(), static_cast<size_t>(sample_rate_hz_ / 100));
// Compute the signal power spectrum.
float downsampled_frame[80];
down_sampler_.DownSample(rtc::ArrayView<const float>(
audio.channels_const_f()[0], audio.num_frames()),
downsampled_frame);
float extended_frame[128];
frame_extender_->ExtendFrame(downsampled_frame, extended_frame);
RemoveDcLevel(extended_frame);
float signal_spectrum[65];
PowerSpectrum(extended_frame, signal_spectrum);
// Classify the signal based on the estimate of the noise spectrum and the
// signal spectrum estimate.
*signal_type = ClassifySignal(signal_spectrum,
noise_spectrum_estimator_.GetNoiseSpectrum(),
data_dumper_);
// Update the noise spectrum based on the signal spectrum.
noise_spectrum_estimator_.Update(signal_spectrum,
initialization_frames_left_ > 0);
// Update the number of frames until a reliable signal spectrum is achieved.
initialization_frames_left_ = std::max(0, initialization_frames_left_ - 1);
if (last_signal_type_ == *signal_type) {
consistent_classification_counter_ =
std::max(0, consistent_classification_counter_ - 1);
} else {
last_signal_type_ = *signal_type;
consistent_classification_counter_ = 3;
}
if (consistent_classification_counter_ > 0) {
*signal_type = SignalClassifier::SignalType::kNonStationary;
}
}
} // namespace webrtc