Revert "Pull the Voice Activity Detector out from the AGC"

This reverts commit 518c683f3e.

Breaks Linux-Asan bot
https://uberchromegw.corp.google.com/i/client.webrtc/builders/Linux%20Asan/builds/4348/steps/libjingle_peerconnection_unittest/logs/stdio

BUG=
TBR=aluebs@webrtc.org

Review URL: https://codereview.webrtc.org/1208793002.

Cr-Commit-Position: refs/heads/master@{#9503}
This commit is contained in:
Bjorn Volcker 2015-06-25 08:46:02 +02:00
parent 518c683f3e
commit 51c7cbb86a
38 changed files with 565 additions and 873 deletions

View File

@ -38,9 +38,17 @@ source_set("audio_processing") {
"aecm/include/echo_control_mobile.h",
"agc/agc.cc",
"agc/agc.h",
"agc/agc_audio_proc.cc",
"agc/agc_audio_proc.h",
"agc/agc_audio_proc_internal.h",
"agc/agc_manager_direct.cc",
"agc/agc_manager_direct.h",
"agc/circular_buffer.cc",
"agc/circular_buffer.h",
"agc/common.h",
"agc/gain_map_internal.h",
"agc/gmm.cc",
"agc/gmm.h",
"agc/histogram.cc",
"agc/histogram.h",
"agc/legacy/analog_agc.c",
@ -48,8 +56,18 @@ source_set("audio_processing") {
"agc/legacy/digital_agc.c",
"agc/legacy/digital_agc.h",
"agc/legacy/gain_control.h",
"agc/noise_gmm_tables.h",
"agc/pitch_based_vad.cc",
"agc/pitch_based_vad.h",
"agc/pitch_internal.cc",
"agc/pitch_internal.h",
"agc/pole_zero_filter.cc",
"agc/pole_zero_filter.h",
"agc/standalone_vad.cc",
"agc/standalone_vad.h",
"agc/utility.cc",
"agc/utility.h",
"agc/voice_gmm_tables.h",
"audio_buffer.cc",
"audio_buffer.h",
"audio_processing_impl.cc",
@ -107,26 +125,6 @@ source_set("audio_processing") {
"utility/delay_estimator_internal.h",
"utility/delay_estimator_wrapper.c",
"utility/delay_estimator_wrapper.h",
"vad/common.h",
"vad/gmm.cc",
"vad/gmm.h",
"vad/noise_gmm_tables.h",
"vad/pitch_based_vad.cc",
"vad/pitch_based_vad.h",
"vad/pitch_internal.cc",
"vad/pitch_internal.h",
"vad/pole_zero_filter.cc",
"vad/pole_zero_filter.h",
"vad/standalone_vad.cc",
"vad/standalone_vad.h",
"vad/vad_audio_proc.cc",
"vad/vad_audio_proc.h",
"vad/vad_audio_proc_internal.h",
"vad/vad_circular_buffer.cc",
"vad/vad_circular_buffer.h",
"vad/voice_activity_detector.cc",
"vad/voice_activity_detector.h",
"vad/voice_gmm_tables.h",
"voice_detection_impl.cc",
"voice_detection_impl.h",
]

View File

@ -14,10 +14,13 @@
#include <cstdlib>
#include <algorithm>
#include <vector>
#include "webrtc/base/checks.h"
#include "webrtc/common_audio/resampler/include/resampler.h"
#include "webrtc/modules/audio_processing/agc/agc_audio_proc.h"
#include "webrtc/modules/audio_processing/agc/common.h"
#include "webrtc/modules/audio_processing/agc/histogram.h"
#include "webrtc/modules/audio_processing/agc/pitch_based_vad.h"
#include "webrtc/modules/audio_processing/agc/standalone_vad.h"
#include "webrtc/modules/audio_processing/agc/utility.h"
#include "webrtc/modules/interface/module_common_types.h"
@ -25,6 +28,7 @@ namespace webrtc {
namespace {
const int kDefaultLevelDbfs = -18;
const double kDefaultVoiceValue = 1.0;
const int kNumAnalysisFrames = 100;
const double kActivityThreshold = 0.3;
@ -32,9 +36,16 @@ const double kActivityThreshold = 0.3;
Agc::Agc()
: target_level_loudness_(Dbfs2Loudness(kDefaultLevelDbfs)),
last_voice_probability_(kDefaultVoiceValue),
target_level_dbfs_(kDefaultLevelDbfs),
standalone_vad_enabled_(true),
histogram_(Histogram::Create(kNumAnalysisFrames)),
inactive_histogram_(Histogram::Create()) {
inactive_histogram_(Histogram::Create()),
audio_processing_(new AgcAudioProc()),
pitch_based_vad_(new PitchBasedVad()),
standalone_vad_(StandaloneVad::Create()),
// Initialize to the most common resampling situation.
resampler_(new Resampler(32000, kSampleRateHz, 1)) {
}
Agc::~Agc() {}
@ -50,13 +61,55 @@ float Agc::AnalyzePreproc(const int16_t* audio, int length) {
}
int Agc::Process(const int16_t* audio, int length, int sample_rate_hz) {
vad_.ProcessChunk(audio, length, sample_rate_hz);
const std::vector<double>& rms = vad_.chunkwise_rms();
const std::vector<double>& probabilities =
vad_.chunkwise_voice_probabilities();
DCHECK_EQ(rms.size(), probabilities.size());
for (size_t i = 0; i < rms.size(); ++i) {
histogram_->Update(rms[i], probabilities[i]);
assert(length == sample_rate_hz / 100);
if (sample_rate_hz > 32000) {
return -1;
}
// Resample to the required rate.
int16_t resampled[kLength10Ms];
const int16_t* resampled_ptr = audio;
if (sample_rate_hz != kSampleRateHz) {
if (resampler_->ResetIfNeeded(sample_rate_hz, kSampleRateHz, 1) != 0) {
return -1;
}
resampler_->Push(audio, length, resampled, kLength10Ms, length);
resampled_ptr = resampled;
}
assert(length == kLength10Ms);
if (standalone_vad_enabled_) {
if (standalone_vad_->AddAudio(resampled_ptr, length) != 0)
return -1;
}
AudioFeatures features;
audio_processing_->ExtractFeatures(resampled_ptr, length, &features);
if (features.num_frames > 0) {
if (features.silence) {
// The other features are invalid, so update the histogram with an
// arbitrary low value.
for (int n = 0; n < features.num_frames; ++n)
histogram_->Update(features.rms[n], 0.01);
return 0;
}
// Initialize to 0.5 which is a neutral value for combining probabilities,
// in case the standalone-VAD is not enabled.
double p_combined[] = {0.5, 0.5, 0.5, 0.5};
static_assert(sizeof(p_combined) / sizeof(p_combined[0]) == kMaxNumFrames,
"combined probability incorrect size");
if (standalone_vad_enabled_) {
if (standalone_vad_->GetActivity(p_combined, kMaxNumFrames) < 0)
return -1;
}
// If any other VAD is enabled it must be combined before calling the
// pitch-based VAD.
if (pitch_based_vad_->VoicingProbability(features, p_combined) < 0)
return -1;
for (int n = 0; n < features.num_frames; n++) {
histogram_->Update(features.rms[n], p_combined[n]);
last_voice_probability_ = p_combined[n];
}
}
return 0;
}
@ -98,4 +151,8 @@ int Agc::set_target_level_dbfs(int level) {
return 0;
}
void Agc::EnableStandaloneVad(bool enable) {
standalone_vad_enabled_ = enable;
}
} // namespace webrtc

View File

@ -12,13 +12,16 @@
#define WEBRTC_MODULES_AUDIO_PROCESSING_AGC_AGC_H_
#include "webrtc/base/scoped_ptr.h"
#include "webrtc/modules/audio_processing/vad/voice_activity_detector.h"
#include "webrtc/typedefs.h"
namespace webrtc {
class AudioFrame;
class AgcAudioProc;
class Histogram;
class PitchBasedVad;
class Resampler;
class StandaloneVad;
class Agc {
public:
@ -41,16 +44,24 @@ class Agc {
virtual int set_target_level_dbfs(int level);
virtual int target_level_dbfs() const { return target_level_dbfs_; }
virtual float voice_probability() const {
return vad_.last_voice_probability();
virtual void EnableStandaloneVad(bool enable);
virtual bool standalone_vad_enabled() const {
return standalone_vad_enabled_;
}
virtual double voice_probability() const { return last_voice_probability_; }
private:
double target_level_loudness_;
double last_voice_probability_;
int target_level_dbfs_;
bool standalone_vad_enabled_;
rtc::scoped_ptr<Histogram> histogram_;
rtc::scoped_ptr<Histogram> inactive_histogram_;
VoiceActivityDetector vad_;
rtc::scoped_ptr<AgcAudioProc> audio_processing_;
rtc::scoped_ptr<PitchBasedVad> pitch_based_vad_;
rtc::scoped_ptr<StandaloneVad> standalone_vad_;
rtc::scoped_ptr<Resampler> resampler_;
};
} // namespace webrtc

View File

@ -8,15 +8,15 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/modules/audio_processing/vad/vad_audio_proc.h"
#include "webrtc/modules/audio_processing/agc/agc_audio_proc.h"
#include <math.h>
#include <stdio.h>
#include "webrtc/common_audio/fft4g.h"
#include "webrtc/modules/audio_processing/vad/vad_audio_proc_internal.h"
#include "webrtc/modules/audio_processing/vad/pitch_internal.h"
#include "webrtc/modules/audio_processing/vad/pole_zero_filter.h"
#include "webrtc/modules/audio_processing/agc/agc_audio_proc_internal.h"
#include "webrtc/modules/audio_processing/agc/pitch_internal.h"
#include "webrtc/modules/audio_processing/agc/pole_zero_filter.h"
extern "C" {
#include "webrtc/modules/audio_coding/codecs/isac/main/source/codec.h"
#include "webrtc/modules/audio_coding/codecs/isac/main/source/lpc_analysis.h"
@ -29,25 +29,23 @@ namespace webrtc {
// The following structures are declared anonymous in iSAC's structs.h. To
// forward declare them, we use this derived class trick.
struct VadAudioProc::PitchAnalysisStruct : public ::PitchAnalysisStruct {};
struct VadAudioProc::PreFiltBankstr : public ::PreFiltBankstr {};
struct AgcAudioProc::PitchAnalysisStruct : public ::PitchAnalysisStruct {};
struct AgcAudioProc::PreFiltBankstr : public ::PreFiltBankstr {};
static const float kFrequencyResolution =
kSampleRateHz / static_cast<float>(VadAudioProc::kDftSize);
static const float kFrequencyResolution = kSampleRateHz /
static_cast<float>(AgcAudioProc::kDftSize);
static const int kSilenceRms = 5;
// TODO(turajs): Make a Create or Init for VadAudioProc.
VadAudioProc::VadAudioProc()
// TODO(turajs): Make a Create or Init for AgcAudioProc.
AgcAudioProc::AgcAudioProc()
: audio_buffer_(),
num_buffer_samples_(kNumPastSignalSamples),
log_old_gain_(-2),
old_lag_(50), // Arbitrary but valid as pitch-lag (in samples).
pitch_analysis_handle_(new PitchAnalysisStruct),
pre_filter_handle_(new PreFiltBankstr),
high_pass_filter_(PoleZeroFilter::Create(kCoeffNumerator,
kFilterOrder,
kCoeffDenominator,
kFilterOrder)) {
high_pass_filter_(PoleZeroFilter::Create(
kCoeffNumerator, kFilterOrder, kCoeffDenominator, kFilterOrder)) {
static_assert(kNumPastSignalSamples + kNumSubframeSamples ==
sizeof(kLpcAnalWin) / sizeof(kLpcAnalWin[0]),
"lpc analysis window incorrect size");
@ -66,16 +64,15 @@ VadAudioProc::VadAudioProc()
WebRtcIsac_InitPitchAnalysis(pitch_analysis_handle_.get());
}
VadAudioProc::~VadAudioProc() {
}
AgcAudioProc::~AgcAudioProc() {}
void VadAudioProc::ResetBuffer() {
void AgcAudioProc::ResetBuffer() {
memcpy(audio_buffer_, &audio_buffer_[kNumSamplesToProcess],
sizeof(audio_buffer_[0]) * kNumPastSignalSamples);
num_buffer_samples_ = kNumPastSignalSamples;
}
int VadAudioProc::ExtractFeatures(const int16_t* frame,
int AgcAudioProc::ExtractFeatures(const int16_t* frame,
int length,
AudioFeatures* features) {
features->num_frames = 0;
@ -88,7 +85,7 @@ int VadAudioProc::ExtractFeatures(const int16_t* frame,
// classification.
if (high_pass_filter_->Filter(frame, kNumSubframeSamples,
&audio_buffer_[num_buffer_samples_]) != 0) {
return -1;
return -1;
}
num_buffer_samples_ += kNumSubframeSamples;
@ -118,8 +115,7 @@ int VadAudioProc::ExtractFeatures(const int16_t* frame,
}
// Computes |kLpcOrder + 1| correlation coefficients.
void VadAudioProc::SubframeCorrelation(double* corr,
int length_corr,
void AgcAudioProc::SubframeCorrelation(double* corr, int length_corr,
int subframe_index) {
assert(length_corr >= kLpcOrder + 1);
double windowed_audio[kNumSubframeSamples + kNumPastSignalSamples];
@ -128,20 +124,20 @@ void VadAudioProc::SubframeCorrelation(double* corr,
for (int n = 0; n < kNumSubframeSamples + kNumPastSignalSamples; n++)
windowed_audio[n] = audio_buffer_[buffer_index++] * kLpcAnalWin[n];
WebRtcIsac_AutoCorr(corr, windowed_audio,
kNumSubframeSamples + kNumPastSignalSamples, kLpcOrder);
WebRtcIsac_AutoCorr(corr, windowed_audio, kNumSubframeSamples +
kNumPastSignalSamples, kLpcOrder);
}
// Compute |kNum10msSubframes| sets of LPC coefficients, one per 10 ms input.
// The analysis window is 15 ms long and it is centered on the first half of
// each 10ms sub-frame. This is equivalent to computing LPC coefficients for the
// first half of each 10 ms subframe.
void VadAudioProc::GetLpcPolynomials(double* lpc, int length_lpc) {
void AgcAudioProc::GetLpcPolynomials(double* lpc, int length_lpc) {
assert(length_lpc >= kNum10msSubframes * (kLpcOrder + 1));
double corr[kLpcOrder + 1];
double reflec_coeff[kLpcOrder];
for (int i = 0, offset_lpc = 0; i < kNum10msSubframes;
i++, offset_lpc += kLpcOrder + 1) {
i++, offset_lpc += kLpcOrder + 1) {
SubframeCorrelation(corr, kLpcOrder + 1, i);
corr[0] *= 1.0001;
// This makes Lev-Durb a bit more stable.
@ -154,8 +150,7 @@ void VadAudioProc::GetLpcPolynomials(double* lpc, int length_lpc) {
// Fit a second order curve to these 3 points and find the location of the
// extremum. The points are inverted before curve fitting.
static float QuadraticInterpolation(float prev_val,
float curr_val,
static float QuadraticInterpolation(float prev_val, float curr_val,
float next_val) {
// Doing the interpolation in |1 / A(z)|^2.
float fractional_index = 0;
@ -163,8 +158,8 @@ static float QuadraticInterpolation(float prev_val,
prev_val = 1.0f / prev_val;
curr_val = 1.0f / curr_val;
fractional_index =
-(next_val - prev_val) * 0.5f / (next_val + prev_val - 2.f * curr_val);
fractional_index = -(next_val - prev_val) * 0.5f / (next_val + prev_val -
2.f * curr_val);
assert(fabs(fractional_index) < 1);
return fractional_index;
}
@ -174,7 +169,7 @@ static float QuadraticInterpolation(float prev_val,
// with the local minimum of A(z). It saves complexity, as we save one
// inversion. Furthermore, we find the first local maximum of magnitude squared,
// to save on one square root.
void VadAudioProc::FindFirstSpectralPeaks(double* f_peak, int length_f_peak) {
void AgcAudioProc::FindFirstSpectralPeaks(double* f_peak, int length_f_peak) {
assert(length_f_peak >= kNum10msSubframes);
double lpc[kNum10msSubframes * (kLpcOrder + 1)];
// For all sub-frames.
@ -198,8 +193,8 @@ void VadAudioProc::FindFirstSpectralPeaks(double* f_peak, int length_f_peak) {
float next_magn_sqr;
bool found_peak = false;
for (int n = 2; n < kNumDftCoefficients - 1; n++) {
next_magn_sqr =
data[2 * n] * data[2 * n] + data[2 * n + 1] * data[2 * n + 1];
next_magn_sqr = data[2 * n] * data[2 * n] +
data[2 * n + 1] * data[2 * n + 1];
if (curr_magn_sqr < prev_magn_sqr && curr_magn_sqr < next_magn_sqr) {
found_peak = true;
index_peak = n - 1;
@ -218,16 +213,15 @@ void VadAudioProc::FindFirstSpectralPeaks(double* f_peak, int length_f_peak) {
} else {
// A peak is found, do a simple quadratic interpolation to get a more
// accurate estimate of the peak location.
fractional_index =
QuadraticInterpolation(prev_magn_sqr, curr_magn_sqr, next_magn_sqr);
fractional_index = QuadraticInterpolation(prev_magn_sqr, curr_magn_sqr,
next_magn_sqr);
}
f_peak[i] = (index_peak + fractional_index) * kFrequencyResolution;
}
}
// Using iSAC functions to estimate pitch gains & lags.
void VadAudioProc::PitchAnalysis(double* log_pitch_gains,
double* pitch_lags_hz,
void AgcAudioProc::PitchAnalysis(double* log_pitch_gains, double* pitch_lags_hz,
int length) {
// TODO(turajs): This can be "imported" from iSAC & and the next two
// constants.
@ -247,27 +241,28 @@ void VadAudioProc::PitchAnalysis(double* log_pitch_gains,
kNumLookaheadSamples];
// Split signal to lower and upper bands
WebRtcIsac_SplitAndFilterFloat(&audio_buffer_[kNumPastSignalSamples], lower,
upper, lower_lookahead, upper_lookahead,
WebRtcIsac_SplitAndFilterFloat(&audio_buffer_[kNumPastSignalSamples],
lower, upper, lower_lookahead, upper_lookahead,
pre_filter_handle_.get());
WebRtcIsac_PitchAnalysis(lower_lookahead, lower_lookahead_pre_filter,
pitch_analysis_handle_.get(), lags, gains);
// Lags are computed on lower-band signal with sampling rate half of the
// input signal.
GetSubframesPitchParameters(
kSampleRateHz / 2, gains, lags, kNumPitchSubframes, kNum10msSubframes,
&log_old_gain_, &old_lag_, log_pitch_gains, pitch_lags_hz);
GetSubframesPitchParameters(kSampleRateHz / 2, gains, lags,
kNumPitchSubframes, kNum10msSubframes,
&log_old_gain_, &old_lag_,
log_pitch_gains, pitch_lags_hz);
}
void VadAudioProc::Rms(double* rms, int length_rms) {
void AgcAudioProc::Rms(double* rms, int length_rms) {
assert(length_rms >= kNum10msSubframes);
int offset = kNumPastSignalSamples;
for (int i = 0; i < kNum10msSubframes; i++) {
rms[i] = 0;
for (int n = 0; n < kNumSubframeSamples; n++, offset++)
rms[i] += audio_buffer_[offset] * audio_buffer_[offset];
rms[i] = sqrt(rms[i] / kNumSubframeSamples);
rms[i] = sqrt(rms[i] / kNumSubframeSamples);
}
}

View File

@ -8,11 +8,11 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_VAD_VAD_AUDIO_PROC_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_VAD_VAD_AUDIO_PROC_H_
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AGC_AGC_AUDIO_PROC_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_AGC_AGC_AUDIO_PROC_H_
#include "webrtc/base/scoped_ptr.h"
#include "webrtc/modules/audio_processing/vad/common.h"
#include "webrtc/modules/audio_processing/agc/common.h"
#include "webrtc/typedefs.h"
namespace webrtc {
@ -20,14 +20,14 @@ namespace webrtc {
class AudioFrame;
class PoleZeroFilter;
class VadAudioProc {
class AgcAudioProc {
public:
// Forward declare iSAC structs.
struct PitchAnalysisStruct;
struct PreFiltBankstr;
VadAudioProc();
~VadAudioProc();
AgcAudioProc();
~AgcAudioProc();
int ExtractFeatures(const int16_t* audio_frame,
int length,
@ -55,8 +55,7 @@ class VadAudioProc {
static const int kNum10msSubframes = 3;
static const int kNumSubframeSamples = kSampleRateHz / 100;
static const int kNumSamplesToProcess =
kNum10msSubframes *
static const int kNumSamplesToProcess = kNum10msSubframes *
kNumSubframeSamples; // Samples in 30 ms @ given sampling rate.
static const int kBufferLength = kNumPastSignalSamples + kNumSamplesToProcess;
static const int kIpLength = kDftSize >> 1;
@ -81,4 +80,4 @@ class VadAudioProc {
} // namespace webrtc
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_VAD_VAD_AUDIO_PROC_H_
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_AGC_AGC_AUDIO_PROC_H_

View File

@ -0,0 +1,81 @@
/*
* Copyright (c) 2012 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.
*/
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AGC_AGC_AUDIO_PROC_INTERNAL_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_AGC_AGC_AUDIO_PROC_INTERNAL_H_
namespace webrtc {
// These values should match MATLAB counterparts for unit-tests to pass.
static const double kCorrWeight[] = {
1.000000, 0.985000, 0.970225, 0.955672, 0.941337, 0.927217, 0.913308,
0.899609, 0.886115, 0.872823, 0.859730, 0.846834, 0.834132, 0.821620,
0.809296, 0.797156, 0.785199
};
static const double kLpcAnalWin[] = {
0.00000000, 0.01314436, 0.02628645, 0.03942400, 0.05255473, 0.06567639,
0.07878670, 0.09188339, 0.10496421, 0.11802689, 0.13106918, 0.14408883,
0.15708358, 0.17005118, 0.18298941, 0.19589602, 0.20876878, 0.22160547,
0.23440387, 0.24716177, 0.25987696, 0.27254725, 0.28517045, 0.29774438,
0.31026687, 0.32273574, 0.33514885, 0.34750406, 0.35979922, 0.37203222,
0.38420093, 0.39630327, 0.40833713, 0.42030043, 0.43219112, 0.44400713,
0.45574642, 0.46740697, 0.47898676, 0.49048379, 0.50189608, 0.51322164,
0.52445853, 0.53560481, 0.54665854, 0.55761782, 0.56848075, 0.57924546,
0.58991008, 0.60047278, 0.61093173, 0.62128512, 0.63153117, 0.64166810,
0.65169416, 0.66160761, 0.67140676, 0.68108990, 0.69065536, 0.70010148,
0.70942664, 0.71862923, 0.72770765, 0.73666033, 0.74548573, 0.75418233,
0.76274862, 0.77118312, 0.77948437, 0.78765094, 0.79568142, 0.80357442,
0.81132858, 0.81894256, 0.82641504, 0.83374472, 0.84093036, 0.84797069,
0.85486451, 0.86161063, 0.86820787, 0.87465511, 0.88095122, 0.88709512,
0.89308574, 0.89892206, 0.90460306, 0.91012776, 0.91549520, 0.92070447,
0.92575465, 0.93064488, 0.93537432, 0.93994213, 0.94434755, 0.94858979,
0.95266814, 0.95658189, 0.96033035, 0.96391289, 0.96732888, 0.97057773,
0.97365889, 0.97657181, 0.97931600, 0.98189099, 0.98429632, 0.98653158,
0.98859639, 0.99049038, 0.99221324, 0.99376466, 0.99514438, 0.99635215,
0.99738778, 0.99825107, 0.99894188, 0.99946010, 0.99980562, 0.99997840,
0.99997840, 0.99980562, 0.99946010, 0.99894188, 0.99825107, 0.99738778,
0.99635215, 0.99514438, 0.99376466, 0.99221324, 0.99049038, 0.98859639,
0.98653158, 0.98429632, 0.98189099, 0.97931600, 0.97657181, 0.97365889,
0.97057773, 0.96732888, 0.96391289, 0.96033035, 0.95658189, 0.95266814,
0.94858979, 0.94434755, 0.93994213, 0.93537432, 0.93064488, 0.92575465,
0.92070447, 0.91549520, 0.91012776, 0.90460306, 0.89892206, 0.89308574,
0.88709512, 0.88095122, 0.87465511, 0.86820787, 0.86161063, 0.85486451,
0.84797069, 0.84093036, 0.83374472, 0.82641504, 0.81894256, 0.81132858,
0.80357442, 0.79568142, 0.78765094, 0.77948437, 0.77118312, 0.76274862,
0.75418233, 0.74548573, 0.73666033, 0.72770765, 0.71862923, 0.70942664,
0.70010148, 0.69065536, 0.68108990, 0.67140676, 0.66160761, 0.65169416,
0.64166810, 0.63153117, 0.62128512, 0.61093173, 0.60047278, 0.58991008,
0.57924546, 0.56848075, 0.55761782, 0.54665854, 0.53560481, 0.52445853,
0.51322164, 0.50189608, 0.49048379, 0.47898676, 0.46740697, 0.45574642,
0.44400713, 0.43219112, 0.42030043, 0.40833713, 0.39630327, 0.38420093,
0.37203222, 0.35979922, 0.34750406, 0.33514885, 0.32273574, 0.31026687,
0.29774438, 0.28517045, 0.27254725, 0.25987696, 0.24716177, 0.23440387,
0.22160547, 0.20876878, 0.19589602, 0.18298941, 0.17005118, 0.15708358,
0.14408883, 0.13106918, 0.11802689, 0.10496421, 0.09188339, 0.07878670,
0.06567639, 0.05255473, 0.03942400, 0.02628645, 0.01314436, 0.00000000
};
static const int kFilterOrder = 2;
static const float kCoeffNumerator[kFilterOrder + 1] = {0.974827f, -1.949650f,
0.974827f};
static const float kCoeffDenominator[kFilterOrder + 1] = {1.0f, -1.971999f,
0.972457f};
static_assert(kFilterOrder + 1 ==
sizeof(kCoeffNumerator) / sizeof(kCoeffNumerator[0]),
"numerator coefficients incorrect size");
static_assert(kFilterOrder + 1 ==
sizeof(kCoeffDenominator) / sizeof(kCoeffDenominator[0]),
"denominator coefficients incorrect size");
} // namespace webrtc
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_AGC_AUDIO_PROCESSING_H_

View File

@ -12,22 +12,20 @@
// routines. However, interpolation of pitch-gain and lags is in a separate
// class and has its own unit-test.
#include "webrtc/modules/audio_processing/vad/vad_audio_proc.h"
#include "webrtc/modules/audio_processing/agc/agc_audio_proc.h"
#include <math.h>
#include <stdio.h>
#include <string>
#include "testing/gtest/include/gtest/gtest.h"
#include "webrtc/modules/audio_processing/vad/common.h"
#include "gtest/gtest.h"
#include "webrtc/modules/audio_processing/agc/common.h"
#include "webrtc/modules/interface/module_common_types.h"
#include "webrtc/test/testsupport/fileutils.h"
namespace webrtc {
TEST(AudioProcessingTest, DISABLED_ComputingFirstSpectralPeak) {
VadAudioProc audioproc;
AgcAudioProc audioproc;
std::string peak_file_name =
test::ResourcePath("audio_processing/agc/agc_spectral_peak", "dat");
@ -41,7 +39,7 @@ TEST(AudioProcessingTest, DISABLED_ComputingFirstSpectralPeak) {
// Read 10 ms audio in each iteration.
const size_t kDataLength = kLength10Ms;
int16_t data[kDataLength] = {0};
int16_t data[kDataLength] = { 0 };
AudioFeatures features;
double sp[kMaxNumFrames];
while (fread(data, sizeof(int16_t), kDataLength, pcm_file) == kDataLength) {

View File

@ -321,7 +321,7 @@ void AgcManagerDirect::SetCaptureMuted(bool muted) {
}
float AgcManagerDirect::voice_probability() {
return agc_->voice_probability();
return static_cast<float>(agc_->voice_probability());
}
int AgcManagerDirect::CheckVolumeAndReset() {

View File

@ -8,44 +8,42 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/modules/audio_processing/vad/vad_circular_buffer.h"
#include "webrtc/modules/audio_processing/agc/circular_buffer.h"
#include <assert.h>
#include <stdlib.h>
namespace webrtc {
VadCircularBuffer::VadCircularBuffer(int buffer_size)
AgcCircularBuffer::AgcCircularBuffer(int buffer_size)
: buffer_(new double[buffer_size]),
is_full_(false),
index_(0),
buffer_size_(buffer_size),
sum_(0) {
}
sum_(0) {}
VadCircularBuffer::~VadCircularBuffer() {
}
AgcCircularBuffer::~AgcCircularBuffer() {}
void VadCircularBuffer::Reset() {
void AgcCircularBuffer::Reset() {
is_full_ = false;
index_ = 0;
sum_ = 0;
}
VadCircularBuffer* VadCircularBuffer::Create(int buffer_size) {
AgcCircularBuffer* AgcCircularBuffer::Create(int buffer_size) {
if (buffer_size <= 0)
return NULL;
return new VadCircularBuffer(buffer_size);
return new AgcCircularBuffer(buffer_size);
}
double VadCircularBuffer::Oldest() const {
double AgcCircularBuffer::Oldest() const {
if (!is_full_)
return buffer_[0];
else
return buffer_[index_];
}
double VadCircularBuffer::Mean() {
double AgcCircularBuffer::Mean() {
double m;
if (is_full_) {
m = sum_ / buffer_size_;
@ -58,7 +56,7 @@ double VadCircularBuffer::Mean() {
return m;
}
void VadCircularBuffer::Insert(double value) {
void AgcCircularBuffer::Insert(double value) {
if (is_full_) {
sum_ -= buffer_[index_];
}
@ -70,13 +68,13 @@ void VadCircularBuffer::Insert(double value) {
index_ = 0;
}
}
int VadCircularBuffer::BufferLevel() {
int AgcCircularBuffer::BufferLevel() {
if (is_full_)
return buffer_size_;
return index_;
}
int VadCircularBuffer::Get(int index, double* value) const {
int AgcCircularBuffer::Get(int index, double* value) const {
int err = ConvertToLinearIndex(&index);
if (err < 0)
return -1;
@ -84,7 +82,7 @@ int VadCircularBuffer::Get(int index, double* value) const {
return 0;
}
int VadCircularBuffer::Set(int index, double value) {
int AgcCircularBuffer::Set(int index, double value) {
int err = ConvertToLinearIndex(&index);
if (err < 0)
return -1;
@ -95,7 +93,7 @@ int VadCircularBuffer::Set(int index, double value) {
return 0;
}
int VadCircularBuffer::ConvertToLinearIndex(int* index) const {
int AgcCircularBuffer::ConvertToLinearIndex(int* index) const {
if (*index < 0 || *index >= buffer_size_)
return -1;
@ -108,7 +106,7 @@ int VadCircularBuffer::ConvertToLinearIndex(int* index) const {
return 0;
}
int VadCircularBuffer::RemoveTransient(int width_threshold,
int AgcCircularBuffer::RemoveTransient(int width_threshold,
double val_threshold) {
if (!is_full_ && index_ < width_threshold + 2)
return 0;

View File

@ -8,8 +8,8 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_VAD_VAD_CIRCULAR_BUFFER_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_VAD_VAD_CIRCULAR_BUFFER_H_
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AGC_CIRCULAR_BUFFER_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_AGC_CIRCULAR_BUFFER_H_
#include "webrtc/base/scoped_ptr.h"
@ -21,10 +21,10 @@ namespace webrtc {
// It is used in class "PitchBasedActivity" to keep track of posterior
// probabilities in the past few seconds. The posterior probabilities are used
// to recursively update prior probabilities.
class VadCircularBuffer {
class AgcCircularBuffer {
public:
static VadCircularBuffer* Create(int buffer_size);
~VadCircularBuffer();
static AgcCircularBuffer* Create(int buffer_size);
~AgcCircularBuffer();
// If buffer is wrapped around.
bool is_full() const { return is_full_; }
@ -44,7 +44,7 @@ class VadCircularBuffer {
int RemoveTransient(int width_threshold, double val_threshold);
private:
explicit VadCircularBuffer(int buffer_size);
explicit AgcCircularBuffer(int buffer_size);
// Get previous values. |index = 0| corresponds to the most recent
// insertion. |index = 1| is the one before the most recent insertion, and
// so on.
@ -66,4 +66,4 @@ class VadCircularBuffer {
};
} // namespace webrtc
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_VAD_VAD_CIRCULAR_BUFFER_H_
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_AGC_CIRCULAR_BUFFER_H_

View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/modules/audio_processing/vad/vad_circular_buffer.h"
#include "webrtc/modules/audio_processing/agc/circular_buffer.h"
#include <stdio.h>
@ -22,7 +22,7 @@ static const double kValThreshold = 1.0;
static const int kLongBuffSize = 100;
static const int kShortBuffSize = 10;
static void InsertSequentially(int k, VadCircularBuffer* circular_buffer) {
static void InsertSequentially(int k, AgcCircularBuffer* circular_buffer) {
double mean_val;
for (int n = 1; n <= k; n++) {
EXPECT_TRUE(!circular_buffer->is_full());
@ -32,20 +32,19 @@ static void InsertSequentially(int k, VadCircularBuffer* circular_buffer) {
}
}
static void Insert(double value,
int num_insertion,
VadCircularBuffer* circular_buffer) {
static void Insert(double value, int num_insertion,
AgcCircularBuffer* circular_buffer) {
for (int n = 0; n < num_insertion; n++)
circular_buffer->Insert(value);
}
static void InsertZeros(int num_zeros, VadCircularBuffer* circular_buffer) {
static void InsertZeros(int num_zeros, AgcCircularBuffer* circular_buffer) {
Insert(0.0, num_zeros, circular_buffer);
}
TEST(VadCircularBufferTest, GeneralTest) {
rtc::scoped_ptr<VadCircularBuffer> circular_buffer(
VadCircularBuffer::Create(kShortBuffSize));
TEST(AgcCircularBufferTest, GeneralTest) {
rtc::scoped_ptr<AgcCircularBuffer> circular_buffer(
AgcCircularBuffer::Create(kShortBuffSize));
double mean_val;
// Mean should return zero if nothing is inserted.
@ -71,9 +70,9 @@ TEST(VadCircularBufferTest, GeneralTest) {
EXPECT_TRUE(circular_buffer->is_full());
}
TEST(VadCircularBufferTest, TransientsRemoval) {
rtc::scoped_ptr<VadCircularBuffer> circular_buffer(
VadCircularBuffer::Create(kLongBuffSize));
TEST(AgcCircularBufferTest, TransientsRemoval) {
rtc::scoped_ptr<AgcCircularBuffer> circular_buffer(
AgcCircularBuffer::Create(kLongBuffSize));
// Let the first transient be in wrap-around.
InsertZeros(kLongBuffSize - kWidthThreshold / 2, circular_buffer.get());
@ -90,9 +89,9 @@ TEST(VadCircularBufferTest, TransientsRemoval) {
}
}
TEST(VadCircularBufferTest, TransientDetection) {
rtc::scoped_ptr<VadCircularBuffer> circular_buffer(
VadCircularBuffer::Create(kLongBuffSize));
TEST(AgcCircularBufferTest, TransientDetection) {
rtc::scoped_ptr<AgcCircularBuffer> circular_buffer(
AgcCircularBuffer::Create(kLongBuffSize));
// Let the first transient be in wrap-around.
int num_insertion = kLongBuffSize - kWidthThreshold / 2;
InsertZeros(num_insertion, circular_buffer.get());
@ -105,8 +104,8 @@ TEST(VadCircularBufferTest, TransientDetection) {
double mean_val = circular_buffer->Mean();
EXPECT_DOUBLE_EQ(num_non_zero_elements * push_val / kLongBuffSize, mean_val);
circular_buffer->Insert(0);
EXPECT_EQ(0,
circular_buffer->RemoveTransient(kWidthThreshold, kValThreshold));
EXPECT_EQ(0, circular_buffer->RemoveTransient(kWidthThreshold,
kValThreshold));
mean_val = circular_buffer->Mean();
EXPECT_DOUBLE_EQ(num_non_zero_elements * push_val / kLongBuffSize, mean_val);
@ -115,8 +114,8 @@ TEST(VadCircularBufferTest, TransientDetection) {
num_insertion = 3;
Insert(push_val, num_insertion, circular_buffer.get());
circular_buffer->Insert(0);
EXPECT_EQ(0,
circular_buffer->RemoveTransient(kWidthThreshold, kValThreshold));
EXPECT_EQ(0, circular_buffer->RemoveTransient(kWidthThreshold,
kValThreshold));
mean_val = circular_buffer->Mean();
EXPECT_DOUBLE_EQ(num_non_zero_elements * push_val / kLongBuffSize, mean_val);
@ -124,8 +123,8 @@ TEST(VadCircularBufferTest, TransientDetection) {
// it shouldn't be considered transient.
Insert(push_val, num_insertion, circular_buffer.get());
num_non_zero_elements += num_insertion;
EXPECT_EQ(0,
circular_buffer->RemoveTransient(kWidthThreshold, kValThreshold));
EXPECT_EQ(0, circular_buffer->RemoveTransient(kWidthThreshold,
kValThreshold));
mean_val = circular_buffer->Mean();
EXPECT_DOUBLE_EQ(num_non_zero_elements * push_val / kLongBuffSize, mean_val);
}

View File

@ -8,8 +8,8 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_VAD_COMMON_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_VAD_COMMON_H_
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AGC_COMMON_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_AGC_COMMON_H_
static const int kSampleRateHz = 16000;
static const int kLength10Ms = kSampleRateHz / 100;
@ -24,4 +24,4 @@ struct AudioFeatures {
bool silence;
};
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_VAD_COMMON_H_
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_AGC_COMMON_H_

View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/modules/audio_processing/vad/gmm.h"
#include "webrtc/modules/audio_processing/agc/gmm.h"
#include <math.h>
#include <stdlib.h>
@ -19,16 +19,13 @@ namespace webrtc {
static const int kMaxDimension = 10;
static void RemoveMean(const double* in,
const double* mean_vec,
int dimension,
double* out) {
static void RemoveMean(const double* in, const double* mean_vec,
int dimension, double* out) {
for (int n = 0; n < dimension; ++n)
out[n] = in[n] - mean_vec[n];
}
static double ComputeExponent(const double* in,
const double* covar_inv,
static double ComputeExponent(const double* in, const double* covar_inv,
int dimension) {
double q = 0;
for (int i = 0; i < dimension; ++i) {
@ -53,7 +50,7 @@ double EvaluateGmm(const double* x, const GmmParameters& gmm_parameters) {
for (int n = 0; n < gmm_parameters.num_mixtures; n++) {
RemoveMean(x, mean_vec, gmm_parameters.dimension, v);
double q = ComputeExponent(v, covar_inv, gmm_parameters.dimension) +
gmm_parameters.weight[n];
gmm_parameters.weight[n];
f += exp(q);
mean_vec += gmm_parameters.dimension;
covar_inv += gmm_parameters.dimension * gmm_parameters.dimension;

View File

@ -8,8 +8,8 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_VAD_GMM_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_VAD_GMM_H_
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AGC_GMM_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_AGC_GMM_H_
namespace webrtc {
@ -42,4 +42,4 @@ struct GmmParameters {
double EvaluateGmm(const double* x, const GmmParameters& gmm_parameters);
} // namespace webrtc
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_VAD_GMM_H_
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_AGC_GMM_H_

View File

@ -8,13 +8,13 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/modules/audio_processing/vad/gmm.h"
#include "webrtc/modules/audio_processing/agc/gmm.h"
#include <math.h>
#include "testing/gtest/include/gtest/gtest.h"
#include "webrtc/modules/audio_processing/vad/noise_gmm_tables.h"
#include "webrtc/modules/audio_processing/vad/voice_gmm_tables.h"
#include "webrtc/modules/audio_processing/agc/noise_gmm_tables.h"
#include "webrtc/modules/audio_processing/agc/voice_gmm_tables.h"
namespace webrtc {

View File

@ -0,0 +1,77 @@
/*
* Copyright (c) 2012 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.
*/
// GMM tables for inactive segments. Generated by MakeGmmTables.m.
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AGC_NOISE_GMM_TABLES_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_AGC_NOISE_GMM_TABLES_H_
static const int kNoiseGmmNumMixtures = 12;
static const int kNoiseGmmDim = 3;
static const double kNoiseGmmCovarInverse[kNoiseGmmNumMixtures]
[kNoiseGmmDim][kNoiseGmmDim] = {
{{ 7.36219567592941e+00, 4.83060785179861e-03, 1.23335151497610e-02},
{ 4.83060785179861e-03, 1.65289507047817e-04, -2.41490588169997e-04},
{ 1.23335151497610e-02, -2.41490588169997e-04, 6.59472060689382e-03}},
{{ 8.70265239309140e+00, -5.30636201431086e-04, 5.44014966585347e-03},
{-5.30636201431086e-04, 3.11095453521008e-04, -1.86287206836035e-04},
{ 5.44014966585347e-03, -1.86287206836035e-04, 6.29493388790744e-04}},
{{ 4.53467851955055e+00, -3.92977536695197e-03, -2.46521420693317e-03},
{-3.92977536695197e-03, 4.94650752632750e-05, -1.08587438501826e-05},
{-2.46521420693317e-03, -1.08587438501826e-05, 9.28793975422261e-05}},
{{ 9.26817997114275e-01, -4.03976069276753e-04, -3.56441427392165e-03},
{-4.03976069276753e-04, 2.51976251631430e-06, 1.46914206734572e-07},
{-3.56441427392165e-03, 1.46914206734572e-07, 8.19914567685373e-05}},
{{ 7.61715986787441e+00, -1.54889041216888e-04, 2.41756280071656e-02},
{-1.54889041216888e-04, 3.50282550461672e-07, -6.27251196972490e-06},
{ 2.41756280071656e-02, -6.27251196972490e-06, 1.45061847649872e-02}},
{{ 8.31193642663158e+00, -3.84070508164323e-04, -3.09750630821876e-02},
{-3.84070508164323e-04, 3.80433432277336e-07, -1.14321142836636e-06},
{-3.09750630821876e-02, -1.14321142836636e-06, 8.35091486289997e-04}},
{{ 9.67283151270894e-01, 5.82465812445039e-05, -3.18350798617053e-03},
{ 5.82465812445039e-05, 2.23762672000318e-07, -7.74196587408623e-07},
{-3.18350798617053e-03, -7.74196587408623e-07, 3.85120938338325e-04}},
{{ 8.28066236985388e+00, 5.87634508319763e-05, 6.99303090891743e-03},
{ 5.87634508319763e-05, 2.93746018618058e-07, 3.40843332882272e-07},
{ 6.99303090891743e-03, 3.40843332882272e-07, 1.99379171190344e-04}},
{{ 6.07488998675646e+00, -1.11494526618473e-02, 5.10013111123381e-03},
{-1.11494526618473e-02, 6.99238879921751e-04, 5.36718550370870e-05},
{ 5.10013111123381e-03, 5.36718550370870e-05, 5.26909853276753e-04}},
{{ 6.90492021419175e+00, 4.20639355257863e-04, -2.38612752336481e-03},
{ 4.20639355257863e-04, 3.31246767338153e-06, -2.42052288150859e-08},
{-2.38612752336481e-03, -2.42052288150859e-08, 4.46608368363412e-04}},
{{ 1.31069150869715e+01, -1.73718583865670e-04, -1.97591814508578e-02},
{-1.73718583865670e-04, 2.80451716300124e-07, 9.96570755379865e-07},
{-1.97591814508578e-02, 9.96570755379865e-07, 2.41361900868847e-03}},
{{ 4.69566344239814e+00, -2.61077567563690e-04, 5.26359000761433e-03},
{-2.61077567563690e-04, 1.82420859823767e-06, -7.83645887541601e-07},
{ 5.26359000761433e-03, -7.83645887541601e-07, 1.33586288288802e-02}}};
static const double kNoiseGmmMean[kNoiseGmmNumMixtures][kNoiseGmmDim] = {
{-2.01386094766163e+00, 1.69702162045397e+02, 7.41715804872181e+01},
{-1.94684591777290e+00, 1.42398396732668e+02, 1.64186321157831e+02},
{-2.29319297562437e+00, 3.86415425589868e+02, 2.13452215267125e+02},
{-3.25487177070268e+00, 1.08668712553616e+03, 2.33119949467419e+02},
{-2.13159632447467e+00, 4.83821702557717e+03, 6.86786166673740e+01},
{-2.26171410780526e+00, 4.79420193982422e+03, 1.53222513286450e+02},
{-3.32166740703185e+00, 4.35161135834358e+03, 1.33206448431316e+02},
{-2.19290322814343e+00, 3.98325506609408e+03, 2.13249167359934e+02},
{-2.02898459255404e+00, 7.37039893155007e+03, 1.12518527491926e+02},
{-2.26150236399500e+00, 1.54896745196145e+03, 1.49717357868579e+02},
{-2.00417668301790e+00, 3.82434760310304e+03, 1.07438913004312e+02},
{-2.30193040814533e+00, 1.43953696546439e+03, 7.04085275122649e+01}};
static const double kNoiseGmmWeights[kNoiseGmmNumMixtures] = {
-1.09422832086193e+01, -1.10847897513425e+01, -1.36767587732187e+01,
-1.79789356118641e+01, -1.42830169160894e+01, -1.56500228061379e+01,
-1.83124990950113e+01, -1.69979436177477e+01, -1.12329424387828e+01,
-1.41311785780639e+01, -1.47171861448585e+01, -1.35963362781839e+01};
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_AGC_NOISE_GMM_TABLES_H_

View File

@ -8,16 +8,16 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/modules/audio_processing/vad/pitch_based_vad.h"
#include "webrtc/modules/audio_processing/agc/pitch_based_vad.h"
#include <assert.h>
#include <math.h>
#include <string.h>
#include "webrtc/modules/audio_processing/vad/vad_circular_buffer.h"
#include "webrtc/modules/audio_processing/vad/common.h"
#include "webrtc/modules/audio_processing/vad/noise_gmm_tables.h"
#include "webrtc/modules/audio_processing/vad/voice_gmm_tables.h"
#include "webrtc/modules/audio_processing/agc/circular_buffer.h"
#include "webrtc/modules/audio_processing/agc/common.h"
#include "webrtc/modules/audio_processing/agc/noise_gmm_tables.h"
#include "webrtc/modules/audio_processing/agc/voice_gmm_tables.h"
#include "webrtc/modules/interface/module_common_types.h"
namespace webrtc {
@ -44,7 +44,7 @@ static double LimitProbability(double p) {
PitchBasedVad::PitchBasedVad()
: p_prior_(kInitialPriorProbability),
circular_buffer_(VadCircularBuffer::Create(kPosteriorHistorySize)) {
circular_buffer_(AgcCircularBuffer::Create(kPosteriorHistorySize)) {
// Setup noise GMM.
noise_gmm_.dimension = kNoiseGmmDim;
noise_gmm_.num_mixtures = kNoiseGmmNumMixtures;
@ -60,8 +60,7 @@ PitchBasedVad::PitchBasedVad()
voice_gmm_.covar_inverse = &kVoiceGmmCovarInverse[0][0][0];
}
PitchBasedVad::~PitchBasedVad() {
}
PitchBasedVad::~PitchBasedVad() {}
int PitchBasedVad::VoicingProbability(const AudioFeatures& features,
double* p_combined) {
@ -91,9 +90,8 @@ int PitchBasedVad::VoicingProbability(const AudioFeatures& features,
pdf_features_given_noise = kEps * pdf_features_given_voice;
}
p = p_prior_ * pdf_features_given_voice /
(pdf_features_given_voice * p_prior_ +
pdf_features_given_noise * (1 - p_prior_));
p = p_prior_ * pdf_features_given_voice / (pdf_features_given_voice *
p_prior_ + pdf_features_given_noise * (1 - p_prior_));
p = LimitProbability(p);

View File

@ -8,18 +8,18 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_VAD_PITCH_BASED_VAD_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_VAD_PITCH_BASED_VAD_H_
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AGC_PITCH_BASED_VAD_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_AGC_PITCH_BASED_VAD_H_
#include "webrtc/base/scoped_ptr.h"
#include "webrtc/modules/audio_processing/vad/common.h"
#include "webrtc/modules/audio_processing/vad/gmm.h"
#include "webrtc/modules/audio_processing/agc/common.h"
#include "webrtc/modules/audio_processing/agc/gmm.h"
#include "webrtc/typedefs.h"
namespace webrtc {
class AudioFrame;
class VadCircularBuffer;
class AgcCircularBuffer;
// Computes the probability of the input audio frame to be active given
// the corresponding pitch-gain and lag of the frame.
@ -37,7 +37,6 @@ class PitchBasedVad {
// then, computes the voicing probabilities and combine them
// with the given values. The result are returned in |p|.
int VoicingProbability(const AudioFeatures& features, double* p_combined);
private:
int UpdatePrior(double p);
@ -50,8 +49,8 @@ class PitchBasedVad {
double p_prior_;
rtc::scoped_ptr<VadCircularBuffer> circular_buffer_;
rtc::scoped_ptr<AgcCircularBuffer> circular_buffer_;
};
} // namespace webrtc
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_VAD_PITCH_BASED_VAD_H_
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_AGC_PITCH_BASED_VAD_H_

View File

@ -8,21 +8,20 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/modules/audio_processing/vad/pitch_based_vad.h"
#include "webrtc/modules/audio_processing/agc/pitch_based_vad.h"
#include <math.h>
#include <stdio.h>
#include <string.h>
#include <string>
#include "testing/gtest/include/gtest/gtest.h"
#include "gtest/gtest.h"
#include "webrtc/test/testsupport/fileutils.h"
namespace webrtc {
TEST(PitchBasedVadTest, VoicingProbabilityTest) {
std::string spectral_peak_file_name =
test::ResourcePath("audio_processing/agc/agc_spectral_peak", "dat");
std::string spectral_peak_file_name = test::ResourcePath(
"audio_processing/agc/agc_spectral_peak", "dat");
FILE* spectral_peak_file = fopen(spectral_peak_file_name.c_str(), "rb");
ASSERT_TRUE(spectral_peak_file != NULL);
@ -52,15 +51,12 @@ TEST(PitchBasedVadTest, VoicingProbabilityTest) {
sizeof(audio_features.spectral_peak[0]), 1,
spectral_peak_file) == 1u) {
double p;
ASSERT_EQ(1u, fread(audio_features.log_pitch_gain,
sizeof(audio_features.log_pitch_gain[0]), 1,
pitch_gain_file));
ASSERT_EQ(1u,
fread(audio_features.pitch_lag_hz,
sizeof(audio_features.pitch_lag_hz[0]), 1, pitch_lag_file));
ASSERT_EQ(1u, fread(&reference_activity_probability,
sizeof(reference_activity_probability), 1,
voicing_prob_file));
ASSERT_EQ(1u, fread(audio_features.log_pitch_gain, sizeof(
audio_features.log_pitch_gain[0]), 1, pitch_gain_file));
ASSERT_EQ(1u, fread(audio_features.pitch_lag_hz, sizeof(
audio_features.pitch_lag_hz[0]), 1, pitch_lag_file));
ASSERT_EQ(1u, fread(&reference_activity_probability, sizeof(
reference_activity_probability), 1, voicing_prob_file));
p = 0.5; // Initialize to the neutral value for combining probabilities.
EXPECT_EQ(0, vad_.VoicingProbability(audio_features, &p));

View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/modules/audio_processing/vad/pitch_internal.h"
#include "webrtc/modules/audio_processing/agc/pitch_internal.h"
#include <cmath>
@ -25,6 +25,7 @@ static void PitchInterpolation(double old_val, const double* in, double* out) {
out[2] = 0.5 * in[2] + 0.5 * in[3];
}
void GetSubframesPitchParameters(int sampling_rate_hz,
double* gains,
double* lags,

View File

@ -8,8 +8,8 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_VAD_PITCH_INTERNAL_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_VAD_PITCH_INTERNAL_H_
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AGC_PITCH_INTERNAL_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_AGC_PITCH_INTERNAL_H_
// TODO(turajs): Write a description of this function. Also be consistent with
// usage of |sampling_rate_hz| vs |kSamplingFreqHz|.
@ -23,4 +23,4 @@ void GetSubframesPitchParameters(int sampling_rate_hz,
double* log_pitch_gain,
double* pitch_lag_hz);
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_VAD_PITCH_INTERNAL_H_
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_AGC_PITCH_INTERNAL_H_

View File

@ -8,11 +8,11 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/modules/audio_processing/vad/pitch_internal.h"
#include "webrtc/modules/audio_processing/agc/pitch_internal.h"
#include <math.h>
#include "testing/gtest/include/gtest/gtest.h"
#include "gtest/gtest.h"
TEST(PitchInternalTest, test) {
const int kSamplingRateHz = 8000;
@ -26,12 +26,12 @@ TEST(PitchInternalTest, test) {
double lags[] = {90, 111, 122, 50};
// Expected outputs
double expected_log_pitch_gain[] = {
-0.541212549898316, -1.45672279045507, -0.80471895621705};
double expected_log_pitch_gain[] = {-0.541212549898316, -1.45672279045507,
-0.80471895621705};
double expected_log_old_gain = log(gains[kNumInputParameters - 1]);
double expected_pitch_lag_hz[] = {
92.3076923076923, 70.9010339734121, 93.0232558139535};
double expected_pitch_lag_hz[] = {92.3076923076923, 70.9010339734121,
93.0232558139535};
double expected_old_lag = lags[kNumInputParameters - 1];
double log_pitch_gain[kNumOutputParameters];

View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/modules/audio_processing/vad/pole_zero_filter.h"
#include "webrtc/modules/audio_processing/agc/pole_zero_filter.h"
#include <stdlib.h>
#include <string.h>
@ -20,10 +20,13 @@ PoleZeroFilter* PoleZeroFilter::Create(const float* numerator_coefficients,
int order_numerator,
const float* denominator_coefficients,
int order_denominator) {
if (order_numerator < 0 || order_denominator < 0 ||
if (order_numerator < 0 ||
order_denominator < 0 ||
order_numerator > kMaxFilterOrder ||
order_denominator > kMaxFilterOrder || denominator_coefficients[0] == 0 ||
numerator_coefficients == NULL || denominator_coefficients == NULL)
order_denominator > kMaxFilterOrder ||
denominator_coefficients[0] == 0 ||
numerator_coefficients == NULL ||
denominator_coefficients == NULL)
return NULL;
return new PoleZeroFilter(numerator_coefficients, order_numerator,
denominator_coefficients, order_denominator);
@ -54,7 +57,8 @@ PoleZeroFilter::PoleZeroFilter(const float* numerator_coefficients,
}
template <typename T>
static float FilterArPast(const T* past, int order, const float* coefficients) {
static float FilterArPast(const T* past, int order,
const float* coefficients) {
float sum = 0.0f;
int past_index = order - 1;
for (int k = 1; k <= order; k++, past_index--)
@ -83,8 +87,8 @@ int PoleZeroFilter::Filter(const int16_t* in,
if (highest_order_ < num_input_samples) {
for (int m = 0; n < num_input_samples; n++, m++) {
output[n] = in[n] * numerator_coefficients_[0];
output[n] +=
FilterArPast(&in[m], order_numerator_, numerator_coefficients_);
output[n] += FilterArPast(&in[m], order_numerator_,
numerator_coefficients_);
output[n] -= FilterArPast(&output[m], order_denominator_,
denominator_coefficients_);
}
@ -95,12 +99,13 @@ int PoleZeroFilter::Filter(const int16_t* in,
sizeof(output[0]) * order_denominator_);
} else {
// Odd case that the length of the input is shorter that filter order.
memmove(past_input_, &past_input_[num_input_samples],
order_numerator_ * sizeof(past_input_[0]));
memmove(past_output_, &past_output_[num_input_samples],
order_denominator_ * sizeof(past_output_[0]));
memmove(past_input_, &past_input_[num_input_samples], order_numerator_ *
sizeof(past_input_[0]));
memmove(past_output_, &past_output_[num_input_samples], order_denominator_ *
sizeof(past_output_[0]));
}
return 0;
}
} // namespace webrtc

View File

@ -8,8 +8,8 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_VAD_POLE_ZERO_FILTER_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_VAD_POLE_ZERO_FILTER_H_
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AGC_POLE_ZERO_FILTER_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_AGC_POLE_ZERO_FILTER_H_
#include "webrtc/typedefs.h"
@ -47,4 +47,4 @@ class PoleZeroFilter {
} // namespace webrtc
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_VAD_POLE_ZERO_FILTER_H_
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_AGC_POLE_ZERO_FILTER_H_

View File

@ -8,49 +8,44 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/modules/audio_processing/vad/pole_zero_filter.h"
#include "webrtc/modules/audio_processing/agc/pole_zero_filter.h"
#include <math.h>
#include <stdio.h>
#include "testing/gtest/include/gtest/gtest.h"
#include "gtest/gtest.h"
#include "webrtc/base/scoped_ptr.h"
#include "webrtc/modules/audio_processing/vad/vad_audio_proc_internal.h"
#include "webrtc/modules/audio_processing/agc/agc_audio_proc_internal.h"
#include "webrtc/test/testsupport/fileutils.h"
namespace webrtc {
static const int kInputSamples = 50;
static const int16_t kInput[kInputSamples] = {
-2136, -7116, 10715, 2464, 3164, 8139, 11393, 24013, -32117, -5544,
-27740, 10181, 14190, -24055, -15912, 17393, 6359, -9950, -13894, 32432,
-23944, 3437, -8381, 19768, 3087, -19795, -5920, 13310, 1407, 3876,
4059, 3524, -23130, 19121, -27900, -24840, 4089, 21422, -3625, 3015,
-11236, 28856, 13424, 6571, -19761, -6361, 15821, -9469, 29727, 32229};
static const int16_t kInput[kInputSamples] = {-2136, -7116, 10715, 2464, 3164,
8139, 11393, 24013, -32117, -5544, -27740, 10181, 14190, -24055, -15912,
17393, 6359, -9950, -13894, 32432, -23944, 3437, -8381, 19768, 3087, -19795,
-5920, 13310, 1407, 3876, 4059, 3524, -23130, 19121, -27900, -24840, 4089,
21422, -3625, 3015, -11236, 28856, 13424, 6571, -19761, -6361, 15821, -9469,
29727, 32229};
static const float kReferenceOutput[kInputSamples] = {
-2082.230472f, -6878.572941f, 10697.090871f, 2358.373952f,
2973.936512f, 7738.580650f, 10690.803213f, 22687.091576f,
-32676.684717f, -5879.621684f, -27359.297432f, 10368.735888f,
13994.584604f, -23676.126249f, -15078.250390f, 17818.253338f,
6577.743123f, -9498.369315f, -13073.651079f, 32460.026588f,
-23391.849347f, 3953.805667f, -7667.761363f, 19995.153447f,
3185.575477f, -19207.365160f, -5143.103201f, 13756.317237f,
1779.654794f, 4142.269755f, 4209.475034f, 3572.991789f,
-22509.089546f, 19307.878964f, -27060.439759f, -23319.042810f,
5547.685267f, 22312.718676f, -2707.309027f, 3852.358490f,
-10135.510093f, 29241.509970f, 13394.397233f, 6340.721417f,
-19510.207905f, -5908.442086f, 15882.301634f, -9211.335255f,
29253.056735f, 30874.443046f};
static const float kReferenceOutput[kInputSamples] = {-2082.230472f,
-6878.572941f, 10697.090871f, 2358.373952f, 2973.936512f, 7738.580650f,
10690.803213f, 22687.091576f, -32676.684717f, -5879.621684f, -27359.297432f,
10368.735888f, 13994.584604f, -23676.126249f, -15078.250390f, 17818.253338f,
6577.743123f, -9498.369315f, -13073.651079f, 32460.026588f, -23391.849347f,
3953.805667f, -7667.761363f, 19995.153447f, 3185.575477f, -19207.365160f,
-5143.103201f, 13756.317237f, 1779.654794f, 4142.269755f, 4209.475034f,
3572.991789f, -22509.089546f, 19307.878964f, -27060.439759f, -23319.042810f,
5547.685267f, 22312.718676f, -2707.309027f, 3852.358490f, -10135.510093f,
29241.509970f, 13394.397233f, 6340.721417f, -19510.207905f, -5908.442086f,
15882.301634f, -9211.335255f, 29253.056735f, 30874.443046f};
class PoleZeroFilterTest : public ::testing::Test {
protected:
PoleZeroFilterTest()
: my_filter_(PoleZeroFilter::Create(kCoeffNumerator,
kFilterOrder,
kCoeffDenominator,
kFilterOrder)) {}
: my_filter_(PoleZeroFilter::Create(
kCoeffNumerator, kFilterOrder, kCoeffDenominator, kFilterOrder)) {}
~PoleZeroFilterTest() {}

View File

@ -8,7 +8,7 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/modules/audio_processing/vad/standalone_vad.h"
#include "webrtc/modules/audio_processing/agc/standalone_vad.h"
#include <assert.h>
@ -21,8 +21,10 @@ namespace webrtc {
static const int kDefaultStandaloneVadMode = 3;
StandaloneVad::StandaloneVad(VadInst* vad)
: vad_(vad), buffer_(), index_(0), mode_(kDefaultStandaloneVadMode) {
}
: vad_(vad),
buffer_(),
index_(0),
mode_(kDefaultStandaloneVadMode) {}
StandaloneVad::~StandaloneVad() {
WebRtcVad_Free(vad_);
@ -91,3 +93,4 @@ int StandaloneVad::set_mode(int mode) {
}
} // namespace webrtc

View File

@ -12,8 +12,8 @@
#define WEBRTC_MODULES_AUDIO_PROCESSING_AGC_STANDALONE_VAD_H_
#include "webrtc/base/scoped_ptr.h"
#include "webrtc/modules/audio_processing/vad/common.h"
#include "webrtc/common_audio/vad/include/webrtc_vad.h"
#include "webrtc/modules/audio_processing/agc/common.h"
#include "webrtc/typedefs.h"
namespace webrtc {

View File

@ -8,11 +8,11 @@
* be found in the AUTHORS file in the root of the source tree.
*/
#include "webrtc/modules/audio_processing/vad/standalone_vad.h"
#include "webrtc/modules/audio_processing/agc/standalone_vad.h"
#include <string.h>
#include "testing/gtest/include/gtest/gtest.h"
#include "gtest/gtest.h"
#include "webrtc/base/scoped_ptr.h"
#include "webrtc/modules/interface/module_common_types.h"
#include "webrtc/test/testsupport/fileutils.h"
@ -22,7 +22,7 @@ namespace webrtc {
TEST(StandaloneVadTest, Api) {
rtc::scoped_ptr<StandaloneVad> vad(StandaloneVad::Create());
int16_t data[kLength10Ms] = {0};
int16_t data[kLength10Ms] = { 0 };
// Valid frame length (for 32 kHz rate), but not what the VAD is expecting.
EXPECT_EQ(-1, vad->AddAudio(data, 320));
@ -58,7 +58,7 @@ TEST(StandaloneVadTest, Api) {
TEST(StandaloneVadTest, DISABLED_ON_IOS(ActivityDetection)) {
rtc::scoped_ptr<StandaloneVad> vad(StandaloneVad::Create());
const size_t kDataLength = kLength10Ms;
int16_t data[kDataLength] = {0};
int16_t data[kDataLength] = { 0 };
FILE* pcm_file =
fopen(test::ResourcePath("audio_processing/agc/agc_audio", "pcm").c_str(),
@ -101,4 +101,4 @@ TEST(StandaloneVadTest, DISABLED_ON_IOS(ActivityDetection)) {
fclose(reference_file);
fclose(pcm_file);
}
} // namespace webrtc
}

View File

@ -0,0 +1,77 @@
/*
* Copyright (c) 2012 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.
*/
// GMM tables for active segments. Generated by MakeGmmTables.m.
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_AGC_VOICE_GMM_TABLES_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_AGC_VOICE_GMM_TABLES_H_
static const int kVoiceGmmNumMixtures = 12;
static const int kVoiceGmmDim = 3;
static const double kVoiceGmmCovarInverse[kVoiceGmmNumMixtures]
[kVoiceGmmDim][kVoiceGmmDim] = {
{{ 1.83673825579513e+00, -8.09791637570095e-04, 4.60106414365986e-03},
{-8.09791637570095e-04, 8.89351738394608e-04, -9.80188953277734e-04},
{ 4.60106414365986e-03, -9.80188953277734e-04, 1.38706060206582e-03}},
{{ 6.76228912850703e+01, -1.98893120119660e-02, -3.53548357253551e-03},
{-1.98893120119660e-02, 3.96216858500530e-05, -4.08492938394097e-05},
{-3.53548357253551e-03, -4.08492938394097e-05, 9.31864352856416e-04}},
{{ 9.98612435944558e+00, -5.27880954316893e-03, -6.30342541619017e-03},
{-5.27880954316893e-03, 4.54359480225226e-05, 6.30804591626044e-05},
{-6.30342541619017e-03, 6.30804591626044e-05, 5.36466441382942e-04}},
{{ 3.39917474216349e+01, -1.56213579433191e-03, -4.01459014990225e-02},
{-1.56213579433191e-03, 6.40415424897724e-05, 6.20076342427833e-05},
{-4.01459014990225e-02, 6.20076342427833e-05, 3.51199070103063e-03}},
{{ 1.34545062271428e+01, -7.94513610147144e-03, -5.34401019341728e-02},
{-7.94513610147144e-03, 1.16511820098649e-04, 4.66063702069293e-05},
{-5.34401019341728e-02, 4.66063702069293e-05, 2.72354323774163e-03}},
{{ 1.08557844314806e+02, -1.54885805673668e-02, -1.88029692674851e-02},
{-1.54885805673668e-02, 1.16404042786406e-04, 6.45579292702802e-06},
{-1.88029692674851e-02, 6.45579292702802e-06, 4.32330478391416e-04}},
{{ 8.22940066541450e+01, -1.15903110231303e-02, -4.92166764865343e-02},
{-1.15903110231303e-02, 7.42510742165261e-05, 3.73007314191290e-06},
{-4.92166764865343e-02, 3.73007314191290e-06, 3.64005221593244e-03}},
{{ 2.31133605685660e+00, -7.83261568950254e-04, 7.45744012346313e-04},
{-7.83261568950254e-04, 1.29460648214142e-05, -2.22774455093730e-06},
{ 7.45744012346313e-04, -2.22774455093730e-06, 1.05117294093010e-04}},
{{ 3.78767849189611e+02, 1.57759761011568e-03, -2.08551217988774e-02},
{ 1.57759761011568e-03, 4.76066236886865e-05, -2.33977412299324e-05},
{-2.08551217988774e-02, -2.33977412299324e-05, 5.24261005371196e-04}},
{{ 6.98580096506135e-01, -5.13850255217378e-04, -4.01124551717056e-04},
{-5.13850255217378e-04, 1.40501021984840e-06, -2.09496928716569e-06},
{-4.01124551717056e-04, -2.09496928716569e-06, 2.82879357740037e-04}},
{{ 2.62770945162399e+00, -2.31825753241430e-03, -5.30447217466318e-03},
{-2.31825753241430e-03, 4.59108572227649e-05, 7.67631886355405e-05},
{-5.30447217466318e-03, 7.67631886355405e-05, 2.28521601674098e-03}},
{{ 1.89940391362152e+02, -4.23280856852379e-03, -2.70608873541399e-02},
{-4.23280856852379e-03, 6.77547582742563e-05, 2.69154203800467e-05},
{-2.70608873541399e-02, 2.69154203800467e-05, 3.88574543373470e-03}}};
static const double kVoiceGmmMean[kVoiceGmmNumMixtures][kVoiceGmmDim] = {
{-2.15020241646536e+00, 4.97079062999877e+02, 4.77078119504505e+02},
{-8.92097680029190e-01, 5.92064964199921e+02, 1.81045145941059e+02},
{-1.29435784144398e+00, 4.98450293410611e+02, 1.71991263804064e+02},
{-1.03925228397884e+00, 4.99511274321571e+02, 1.05838336539105e+02},
{-1.29229047206129e+00, 4.15026762566707e+02, 1.12861119017125e+02},
{-7.88748114599810e-01, 4.48739336688113e+02, 1.89784216956337e+02},
{-8.77777402332642e-01, 4.86620285054533e+02, 1.13477708016491e+02},
{-2.06465957063057e+00, 6.33385049870607e+02, 2.32758546796149e+02},
{-6.98893789231685e-01, 5.93622051503385e+02, 1.92536982473203e+02},
{-2.55901217508894e+00, 1.55914919756205e+03, 1.39769980835570e+02},
{-1.92070024165837e+00, 4.87983940444185e+02, 1.02745468128289e+02},
{-7.29187507662854e-01, 5.22717685022855e+02, 1.16377942283991e+02}};
static const double kVoiceGmmWeights[kVoiceGmmNumMixtures] = {
-1.39789694361035e+01, -1.19527720202104e+01, -1.32396317929055e+01,
-1.09436815209238e+01, -1.13440027478149e+01, -1.12200721834504e+01,
-1.02537324043693e+01, -1.60789861938302e+01, -1.03394494048344e+01,
-1.83207938586818e+01, -1.31186044948288e+01, -9.52479998673554e+00};
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_AGC_VOICE_GMM_TABLES_H_

View File

@ -48,9 +48,17 @@
'aecm/include/echo_control_mobile.h',
'agc/agc.cc',
'agc/agc.h',
'agc/agc_audio_proc.cc',
'agc/agc_audio_proc.h',
'agc/agc_audio_proc_internal.h',
'agc/agc_manager_direct.cc',
'agc/agc_manager_direct.h',
'agc/circular_buffer.cc',
'agc/circular_buffer.h',
'agc/common.h',
'agc/gain_map_internal.h',
'agc/gmm.cc',
'agc/gmm.h',
'agc/histogram.cc',
'agc/histogram.h',
'agc/legacy/analog_agc.c',
@ -58,8 +66,18 @@
'agc/legacy/digital_agc.c',
'agc/legacy/digital_agc.h',
'agc/legacy/gain_control.h',
'agc/noise_gmm_tables.h',
'agc/pitch_based_vad.cc',
'agc/pitch_based_vad.h',
'agc/pitch_internal.cc',
'agc/pitch_internal.h',
'agc/pole_zero_filter.cc',
'agc/pole_zero_filter.h',
'agc/standalone_vad.cc',
'agc/standalone_vad.h',
'agc/utility.cc',
'agc/utility.h',
'agc/voice_gmm_tables.h',
'audio_buffer.cc',
'audio_buffer.h',
'audio_processing_impl.cc',
@ -117,26 +135,6 @@
'utility/delay_estimator_internal.h',
'utility/delay_estimator_wrapper.c',
'utility/delay_estimator_wrapper.h',
'vad/common.h',
'vad/gmm.cc',
'vad/gmm.h',
'vad/noise_gmm_tables.h',
'vad/pitch_based_vad.cc',
'vad/pitch_based_vad.h',
'vad/pitch_internal.cc',
'vad/pitch_internal.h',
'vad/pole_zero_filter.cc',
'vad/pole_zero_filter.h',
'vad/standalone_vad.cc',
'vad/standalone_vad.h',
'vad/vad_audio_proc.cc',
'vad/vad_audio_proc.h',
'vad/vad_audio_proc_internal.h',
'vad/vad_circular_buffer.cc',
'vad/vad_circular_buffer.h',
'vad/voice_activity_detector.cc',
'vad/voice_activity_detector.h',
'vad/voice_gmm_tables.h',
'voice_detection_impl.cc',
'voice_detection_impl.h',
],

View File

@ -1,85 +0,0 @@
/*
* Copyright (c) 2012 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.
*/
// GMM tables for inactive segments. Generated by MakeGmmTables.m.
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_VAD_NOISE_GMM_TABLES_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_VAD_NOISE_GMM_TABLES_H_
static const int kNoiseGmmNumMixtures = 12;
static const int kNoiseGmmDim = 3;
static const double
kNoiseGmmCovarInverse[kNoiseGmmNumMixtures][kNoiseGmmDim][kNoiseGmmDim] = {
{{7.36219567592941e+00, 4.83060785179861e-03, 1.23335151497610e-02},
{4.83060785179861e-03, 1.65289507047817e-04, -2.41490588169997e-04},
{1.23335151497610e-02, -2.41490588169997e-04, 6.59472060689382e-03}},
{{8.70265239309140e+00, -5.30636201431086e-04, 5.44014966585347e-03},
{-5.30636201431086e-04, 3.11095453521008e-04, -1.86287206836035e-04},
{5.44014966585347e-03, -1.86287206836035e-04, 6.29493388790744e-04}},
{{4.53467851955055e+00, -3.92977536695197e-03, -2.46521420693317e-03},
{-3.92977536695197e-03, 4.94650752632750e-05, -1.08587438501826e-05},
{-2.46521420693317e-03, -1.08587438501826e-05, 9.28793975422261e-05}},
{{9.26817997114275e-01, -4.03976069276753e-04, -3.56441427392165e-03},
{-4.03976069276753e-04, 2.51976251631430e-06, 1.46914206734572e-07},
{-3.56441427392165e-03, 1.46914206734572e-07, 8.19914567685373e-05}},
{{7.61715986787441e+00, -1.54889041216888e-04, 2.41756280071656e-02},
{-1.54889041216888e-04, 3.50282550461672e-07, -6.27251196972490e-06},
{2.41756280071656e-02, -6.27251196972490e-06, 1.45061847649872e-02}},
{{8.31193642663158e+00, -3.84070508164323e-04, -3.09750630821876e-02},
{-3.84070508164323e-04, 3.80433432277336e-07, -1.14321142836636e-06},
{-3.09750630821876e-02, -1.14321142836636e-06, 8.35091486289997e-04}},
{{9.67283151270894e-01, 5.82465812445039e-05, -3.18350798617053e-03},
{5.82465812445039e-05, 2.23762672000318e-07, -7.74196587408623e-07},
{-3.18350798617053e-03, -7.74196587408623e-07, 3.85120938338325e-04}},
{{8.28066236985388e+00, 5.87634508319763e-05, 6.99303090891743e-03},
{5.87634508319763e-05, 2.93746018618058e-07, 3.40843332882272e-07},
{6.99303090891743e-03, 3.40843332882272e-07, 1.99379171190344e-04}},
{{6.07488998675646e+00, -1.11494526618473e-02, 5.10013111123381e-03},
{-1.11494526618473e-02, 6.99238879921751e-04, 5.36718550370870e-05},
{5.10013111123381e-03, 5.36718550370870e-05, 5.26909853276753e-04}},
{{6.90492021419175e+00, 4.20639355257863e-04, -2.38612752336481e-03},
{4.20639355257863e-04, 3.31246767338153e-06, -2.42052288150859e-08},
{-2.38612752336481e-03, -2.42052288150859e-08, 4.46608368363412e-04}},
{{1.31069150869715e+01, -1.73718583865670e-04, -1.97591814508578e-02},
{-1.73718583865670e-04, 2.80451716300124e-07, 9.96570755379865e-07},
{-1.97591814508578e-02, 9.96570755379865e-07, 2.41361900868847e-03}},
{{4.69566344239814e+00, -2.61077567563690e-04, 5.26359000761433e-03},
{-2.61077567563690e-04, 1.82420859823767e-06, -7.83645887541601e-07},
{5.26359000761433e-03, -7.83645887541601e-07, 1.33586288288802e-02}}};
static const double kNoiseGmmMean[kNoiseGmmNumMixtures][kNoiseGmmDim] = {
{-2.01386094766163e+00, 1.69702162045397e+02, 7.41715804872181e+01},
{-1.94684591777290e+00, 1.42398396732668e+02, 1.64186321157831e+02},
{-2.29319297562437e+00, 3.86415425589868e+02, 2.13452215267125e+02},
{-3.25487177070268e+00, 1.08668712553616e+03, 2.33119949467419e+02},
{-2.13159632447467e+00, 4.83821702557717e+03, 6.86786166673740e+01},
{-2.26171410780526e+00, 4.79420193982422e+03, 1.53222513286450e+02},
{-3.32166740703185e+00, 4.35161135834358e+03, 1.33206448431316e+02},
{-2.19290322814343e+00, 3.98325506609408e+03, 2.13249167359934e+02},
{-2.02898459255404e+00, 7.37039893155007e+03, 1.12518527491926e+02},
{-2.26150236399500e+00, 1.54896745196145e+03, 1.49717357868579e+02},
{-2.00417668301790e+00, 3.82434760310304e+03, 1.07438913004312e+02},
{-2.30193040814533e+00, 1.43953696546439e+03, 7.04085275122649e+01}};
static const double kNoiseGmmWeights[kNoiseGmmNumMixtures] = {
-1.09422832086193e+01,
-1.10847897513425e+01,
-1.36767587732187e+01,
-1.79789356118641e+01,
-1.42830169160894e+01,
-1.56500228061379e+01,
-1.83124990950113e+01,
-1.69979436177477e+01,
-1.12329424387828e+01,
-1.41311785780639e+01,
-1.47171861448585e+01,
-1.35963362781839e+01};
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_VAD_NOISE_GMM_TABLES_H_

View File

@ -1,94 +0,0 @@
/*
* Copyright (c) 2012 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.
*/
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_VAD_VAD_AUDIO_PROC_INTERNAL_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_VAD_VAD_AUDIO_PROC_INTERNAL_H_
namespace webrtc {
// These values should match MATLAB counterparts for unit-tests to pass.
static const double kCorrWeight[] = {1.000000,
0.985000,
0.970225,
0.955672,
0.941337,
0.927217,
0.913308,
0.899609,
0.886115,
0.872823,
0.859730,
0.846834,
0.834132,
0.821620,
0.809296,
0.797156,
0.785199};
static const double kLpcAnalWin[] = {
0.00000000, 0.01314436, 0.02628645, 0.03942400, 0.05255473, 0.06567639,
0.07878670, 0.09188339, 0.10496421, 0.11802689, 0.13106918, 0.14408883,
0.15708358, 0.17005118, 0.18298941, 0.19589602, 0.20876878, 0.22160547,
0.23440387, 0.24716177, 0.25987696, 0.27254725, 0.28517045, 0.29774438,
0.31026687, 0.32273574, 0.33514885, 0.34750406, 0.35979922, 0.37203222,
0.38420093, 0.39630327, 0.40833713, 0.42030043, 0.43219112, 0.44400713,
0.45574642, 0.46740697, 0.47898676, 0.49048379, 0.50189608, 0.51322164,
0.52445853, 0.53560481, 0.54665854, 0.55761782, 0.56848075, 0.57924546,
0.58991008, 0.60047278, 0.61093173, 0.62128512, 0.63153117, 0.64166810,
0.65169416, 0.66160761, 0.67140676, 0.68108990, 0.69065536, 0.70010148,
0.70942664, 0.71862923, 0.72770765, 0.73666033, 0.74548573, 0.75418233,
0.76274862, 0.77118312, 0.77948437, 0.78765094, 0.79568142, 0.80357442,
0.81132858, 0.81894256, 0.82641504, 0.83374472, 0.84093036, 0.84797069,
0.85486451, 0.86161063, 0.86820787, 0.87465511, 0.88095122, 0.88709512,
0.89308574, 0.89892206, 0.90460306, 0.91012776, 0.91549520, 0.92070447,
0.92575465, 0.93064488, 0.93537432, 0.93994213, 0.94434755, 0.94858979,
0.95266814, 0.95658189, 0.96033035, 0.96391289, 0.96732888, 0.97057773,
0.97365889, 0.97657181, 0.97931600, 0.98189099, 0.98429632, 0.98653158,
0.98859639, 0.99049038, 0.99221324, 0.99376466, 0.99514438, 0.99635215,
0.99738778, 0.99825107, 0.99894188, 0.99946010, 0.99980562, 0.99997840,
0.99997840, 0.99980562, 0.99946010, 0.99894188, 0.99825107, 0.99738778,
0.99635215, 0.99514438, 0.99376466, 0.99221324, 0.99049038, 0.98859639,
0.98653158, 0.98429632, 0.98189099, 0.97931600, 0.97657181, 0.97365889,
0.97057773, 0.96732888, 0.96391289, 0.96033035, 0.95658189, 0.95266814,
0.94858979, 0.94434755, 0.93994213, 0.93537432, 0.93064488, 0.92575465,
0.92070447, 0.91549520, 0.91012776, 0.90460306, 0.89892206, 0.89308574,
0.88709512, 0.88095122, 0.87465511, 0.86820787, 0.86161063, 0.85486451,
0.84797069, 0.84093036, 0.83374472, 0.82641504, 0.81894256, 0.81132858,
0.80357442, 0.79568142, 0.78765094, 0.77948437, 0.77118312, 0.76274862,
0.75418233, 0.74548573, 0.73666033, 0.72770765, 0.71862923, 0.70942664,
0.70010148, 0.69065536, 0.68108990, 0.67140676, 0.66160761, 0.65169416,
0.64166810, 0.63153117, 0.62128512, 0.61093173, 0.60047278, 0.58991008,
0.57924546, 0.56848075, 0.55761782, 0.54665854, 0.53560481, 0.52445853,
0.51322164, 0.50189608, 0.49048379, 0.47898676, 0.46740697, 0.45574642,
0.44400713, 0.43219112, 0.42030043, 0.40833713, 0.39630327, 0.38420093,
0.37203222, 0.35979922, 0.34750406, 0.33514885, 0.32273574, 0.31026687,
0.29774438, 0.28517045, 0.27254725, 0.25987696, 0.24716177, 0.23440387,
0.22160547, 0.20876878, 0.19589602, 0.18298941, 0.17005118, 0.15708358,
0.14408883, 0.13106918, 0.11802689, 0.10496421, 0.09188339, 0.07878670,
0.06567639, 0.05255473, 0.03942400, 0.02628645, 0.01314436, 0.00000000};
static const int kFilterOrder = 2;
static const float kCoeffNumerator[kFilterOrder + 1] = {0.974827f,
-1.949650f,
0.974827f};
static const float kCoeffDenominator[kFilterOrder + 1] = {1.0f,
-1.971999f,
0.972457f};
static_assert(kFilterOrder + 1 ==
sizeof(kCoeffNumerator) / sizeof(kCoeffNumerator[0]),
"numerator coefficients incorrect size");
static_assert(kFilterOrder + 1 ==
sizeof(kCoeffDenominator) / sizeof(kCoeffDenominator[0]),
"denominator coefficients incorrect size");
} // namespace webrtc
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_VAD_VAD_AUDIO_PROCESSING_H_

View File

@ -1,87 +0,0 @@
/*
* Copyright (c) 2015 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/vad/voice_activity_detector.h"
#include <algorithm>
#include "webrtc/base/checks.h"
namespace webrtc {
namespace {
const int kMaxLength = 320;
const int kNumChannels = 1;
const double kDefaultVoiceValue = 1.0;
const double kNeutralProbability = 0.5;
const double kLowProbability = 0.01;
} // namespace
VoiceActivityDetector::VoiceActivityDetector()
: last_voice_probability_(kDefaultVoiceValue),
// Initialize to the most common resampling situation.
resampler_(kMaxLength, kLength10Ms, kNumChannels),
standalone_vad_(StandaloneVad::Create()) {
}
// Because ISAC has a different chunk length, it updates
// |chunkwise_voice_probabilities_| and |chunkwise_rms_| when there is new data.
// Otherwise it clears them.
void VoiceActivityDetector::ProcessChunk(const int16_t* audio,
int length,
int sample_rate_hz) {
DCHECK_EQ(length, sample_rate_hz / 100);
DCHECK_LE(length, kMaxLength);
// Resample to the required rate.
const int16_t* resampled_ptr = audio;
if (sample_rate_hz != kSampleRateHz) {
CHECK_EQ(
resampler_.ResetIfNeeded(sample_rate_hz, kSampleRateHz, kNumChannels),
0);
resampler_.Push(audio, length, resampled_, kLength10Ms, length);
resampled_ptr = resampled_;
}
DCHECK_EQ(length, kLength10Ms);
// Each chunk needs to be passed into |standalone_vad_|, because internally it
// buffers the audio and processes it all at once when GetActivity() is
// called.
CHECK_EQ(standalone_vad_->AddAudio(audio, length), 0);
audio_processing_.ExtractFeatures(resampled_ptr, length, &features_);
chunkwise_voice_probabilities_.resize(features_.num_frames);
chunkwise_rms_.resize(features_.num_frames);
std::copy(features_.rms, features_.rms + chunkwise_rms_.size(),
chunkwise_rms_.begin());
if (features_.num_frames > 0) {
if (features_.silence) {
// The other features are invalid, so set the voice probabilities to an
// arbitrary low value.
std::fill(chunkwise_voice_probabilities_.begin(),
chunkwise_voice_probabilities_.end(), kLowProbability);
} else {
std::fill(chunkwise_voice_probabilities_.begin(),
chunkwise_voice_probabilities_.end(), kNeutralProbability);
CHECK_GE(
standalone_vad_->GetActivity(&chunkwise_voice_probabilities_[0],
chunkwise_voice_probabilities_.size()),
0);
CHECK_GE(pitch_based_vad_.VoicingProbability(
features_, &chunkwise_voice_probabilities_[0]),
0);
}
last_voice_probability_ = chunkwise_voice_probabilities_.back();
}
}
} // namespace webrtc

View File

@ -1,70 +0,0 @@
/*
* Copyright (c) 2015 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.
*/
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_VAD_VOICE_ACTIVITY_DETECTOR_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_VAD_VOICE_ACTIVITY_DETECTOR_H_
#include <vector>
#include "webrtc/base/scoped_ptr.h"
#include "webrtc/common_audio/resampler/include/resampler.h"
#include "webrtc/modules/audio_processing/vad/vad_audio_proc.h"
#include "webrtc/modules/audio_processing/vad/common.h"
#include "webrtc/modules/audio_processing/vad/pitch_based_vad.h"
#include "webrtc/modules/audio_processing/vad/standalone_vad.h"
namespace webrtc {
// A Voice Activity Detector (VAD) that combines the voice probability from the
// StandaloneVad and PitchBasedVad to get a more robust estimation.
class VoiceActivityDetector {
public:
VoiceActivityDetector();
// Processes each audio chunk and estimates the voice probability. The maximum
// supported sample rate is 32kHz.
// TODO(aluebs): Change |length| to size_t.
void ProcessChunk(const int16_t* audio, int length, int sample_rate_hz);
// Returns a vector of voice probabilities for each chunk. It can be empty for
// some chunks, but it catches up afterwards returning multiple values at
// once.
const std::vector<double>& chunkwise_voice_probabilities() const {
return chunkwise_voice_probabilities_;
}
// Returns a vector of RMS values for each chunk. It has the same length as
// chunkwise_voice_probabilities().
const std::vector<double>& chunkwise_rms() const { return chunkwise_rms_; }
// Returns the last voice probability, regardless of the internal
// implementation, although it has a few chunks of delay.
float last_voice_probability() const { return last_voice_probability_; }
private:
// TODO(aluebs): Change these to float.
std::vector<double> chunkwise_voice_probabilities_;
std::vector<double> chunkwise_rms_;
float last_voice_probability_;
Resampler resampler_;
VadAudioProc audio_processing_;
rtc::scoped_ptr<StandaloneVad> standalone_vad_;
PitchBasedVad pitch_based_vad_;
int16_t resampled_[kLength10Ms];
AudioFeatures features_;
};
} // namespace webrtc
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_VAD_VOICE_ACTIVITY_DETECTOR_H_

View File

@ -1,168 +0,0 @@
/*
* Copyright (c) 2015 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/vad/voice_activity_detector.h"
#include <algorithm>
#include <vector>
#include "testing/gtest/include/gtest/gtest.h"
#include "webrtc/test/testsupport/fileutils.h"
namespace webrtc {
namespace {
const int kStartTimeSec = 16;
const float kMeanSpeechProbability = 0.3f;
const float kMaxNoiseProbability = 0.1f;
const size_t kNumChunks = 300u;
const size_t kNumChunksPerIsacBlock = 3;
void GenerateNoise(std::vector<int16_t>* data) {
for (size_t i = 0; i < data->size(); ++i) {
// std::rand returns between 0 and RAND_MAX, but this will work because it
// wraps into some random place.
(*data)[i] = std::rand();
}
}
} // namespace
TEST(VoiceActivityDetectorTest, ConstructorSetsDefaultValues) {
const float kDefaultVoiceValue = 1.f;
VoiceActivityDetector vad;
std::vector<double> p = vad.chunkwise_voice_probabilities();
std::vector<double> rms = vad.chunkwise_rms();
EXPECT_EQ(p.size(), 0u);
EXPECT_EQ(rms.size(), 0u);
EXPECT_FLOAT_EQ(vad.last_voice_probability(), kDefaultVoiceValue);
}
TEST(VoiceActivityDetectorTest, Speech16kHzHasHighVoiceProbabilities) {
const int kSampleRateHz = 16000;
const int kLength10Ms = kSampleRateHz / 100;
VoiceActivityDetector vad;
std::vector<int16_t> data(kLength10Ms);
float mean_probability = 0.f;
FILE* pcm_file =
fopen(test::ResourcePath("audio_processing/transient/audio16kHz", "pcm")
.c_str(),
"rb");
ASSERT_TRUE(pcm_file != nullptr);
// The silences in the file are skipped to get a more robust voice probability
// for speech.
ASSERT_EQ(fseek(pcm_file, kStartTimeSec * kSampleRateHz * sizeof(data[0]),
SEEK_SET),
0);
size_t num_chunks = 0;
while (fread(&data[0], sizeof(data[0]), data.size(), pcm_file) ==
data.size()) {
vad.ProcessChunk(&data[0], data.size(), kSampleRateHz);
mean_probability += vad.last_voice_probability();
++num_chunks;
}
mean_probability /= num_chunks;
EXPECT_GT(mean_probability, kMeanSpeechProbability);
}
TEST(VoiceActivityDetectorTest, Speech32kHzHasHighVoiceProbabilities) {
const int kSampleRateHz = 32000;
const int kLength10Ms = kSampleRateHz / 100;
VoiceActivityDetector vad;
std::vector<int16_t> data(kLength10Ms);
float mean_probability = 0.f;
FILE* pcm_file =
fopen(test::ResourcePath("audio_processing/transient/audio32kHz", "pcm")
.c_str(),
"rb");
ASSERT_TRUE(pcm_file != nullptr);
// The silences in the file are skipped to get a more robust voice probability
// for speech.
ASSERT_EQ(fseek(pcm_file, kStartTimeSec * kSampleRateHz * sizeof(data[0]),
SEEK_SET),
0);
size_t num_chunks = 0;
while (fread(&data[0], sizeof(data[0]), data.size(), pcm_file) ==
data.size()) {
vad.ProcessChunk(&data[0], data.size(), kSampleRateHz);
mean_probability += vad.last_voice_probability();
++num_chunks;
}
mean_probability /= num_chunks;
EXPECT_GT(mean_probability, kMeanSpeechProbability);
}
TEST(VoiceActivityDetectorTest, Noise16kHzHasLowVoiceProbabilities) {
VoiceActivityDetector vad;
std::vector<int16_t> data(kLength10Ms);
float max_probability = 0.f;
std::srand(42);
for (size_t i = 0; i < kNumChunks; ++i) {
GenerateNoise(&data);
vad.ProcessChunk(&data[0], data.size(), kSampleRateHz);
// Before the |vad has enough data to process an ISAC block it will return
// the default value, 1.f, which would ruin the |max_probability| value.
if (i > kNumChunksPerIsacBlock) {
max_probability = std::max(max_probability, vad.last_voice_probability());
}
}
EXPECT_LT(max_probability, kMaxNoiseProbability);
}
TEST(VoiceActivityDetectorTest, Noise32kHzHasLowVoiceProbabilities) {
VoiceActivityDetector vad;
std::vector<int16_t> data(2 * kLength10Ms);
float max_probability = 0.f;
std::srand(42);
for (size_t i = 0; i < kNumChunks; ++i) {
GenerateNoise(&data);
vad.ProcessChunk(&data[0], data.size(), 2 * kSampleRateHz);
// Before the |vad has enough data to process an ISAC block it will return
// the default value, 1.f, which would ruin the |max_probability| value.
if (i > kNumChunksPerIsacBlock) {
max_probability = std::max(max_probability, vad.last_voice_probability());
}
}
EXPECT_LT(max_probability, kMaxNoiseProbability);
}
} // namespace webrtc

View File

@ -1,85 +0,0 @@
/*
* Copyright (c) 2012 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.
*/
// GMM tables for active segments. Generated by MakeGmmTables.m.
#ifndef WEBRTC_MODULES_AUDIO_PROCESSING_VAD_VOICE_GMM_TABLES_H_
#define WEBRTC_MODULES_AUDIO_PROCESSING_VAD_VOICE_GMM_TABLES_H_
static const int kVoiceGmmNumMixtures = 12;
static const int kVoiceGmmDim = 3;
static const double
kVoiceGmmCovarInverse[kVoiceGmmNumMixtures][kVoiceGmmDim][kVoiceGmmDim] = {
{{1.83673825579513e+00, -8.09791637570095e-04, 4.60106414365986e-03},
{-8.09791637570095e-04, 8.89351738394608e-04, -9.80188953277734e-04},
{4.60106414365986e-03, -9.80188953277734e-04, 1.38706060206582e-03}},
{{6.76228912850703e+01, -1.98893120119660e-02, -3.53548357253551e-03},
{-1.98893120119660e-02, 3.96216858500530e-05, -4.08492938394097e-05},
{-3.53548357253551e-03, -4.08492938394097e-05, 9.31864352856416e-04}},
{{9.98612435944558e+00, -5.27880954316893e-03, -6.30342541619017e-03},
{-5.27880954316893e-03, 4.54359480225226e-05, 6.30804591626044e-05},
{-6.30342541619017e-03, 6.30804591626044e-05, 5.36466441382942e-04}},
{{3.39917474216349e+01, -1.56213579433191e-03, -4.01459014990225e-02},
{-1.56213579433191e-03, 6.40415424897724e-05, 6.20076342427833e-05},
{-4.01459014990225e-02, 6.20076342427833e-05, 3.51199070103063e-03}},
{{1.34545062271428e+01, -7.94513610147144e-03, -5.34401019341728e-02},
{-7.94513610147144e-03, 1.16511820098649e-04, 4.66063702069293e-05},
{-5.34401019341728e-02, 4.66063702069293e-05, 2.72354323774163e-03}},
{{1.08557844314806e+02, -1.54885805673668e-02, -1.88029692674851e-02},
{-1.54885805673668e-02, 1.16404042786406e-04, 6.45579292702802e-06},
{-1.88029692674851e-02, 6.45579292702802e-06, 4.32330478391416e-04}},
{{8.22940066541450e+01, -1.15903110231303e-02, -4.92166764865343e-02},
{-1.15903110231303e-02, 7.42510742165261e-05, 3.73007314191290e-06},
{-4.92166764865343e-02, 3.73007314191290e-06, 3.64005221593244e-03}},
{{2.31133605685660e+00, -7.83261568950254e-04, 7.45744012346313e-04},
{-7.83261568950254e-04, 1.29460648214142e-05, -2.22774455093730e-06},
{7.45744012346313e-04, -2.22774455093730e-06, 1.05117294093010e-04}},
{{3.78767849189611e+02, 1.57759761011568e-03, -2.08551217988774e-02},
{1.57759761011568e-03, 4.76066236886865e-05, -2.33977412299324e-05},
{-2.08551217988774e-02, -2.33977412299324e-05, 5.24261005371196e-04}},
{{6.98580096506135e-01, -5.13850255217378e-04, -4.01124551717056e-04},
{-5.13850255217378e-04, 1.40501021984840e-06, -2.09496928716569e-06},
{-4.01124551717056e-04, -2.09496928716569e-06, 2.82879357740037e-04}},
{{2.62770945162399e+00, -2.31825753241430e-03, -5.30447217466318e-03},
{-2.31825753241430e-03, 4.59108572227649e-05, 7.67631886355405e-05},
{-5.30447217466318e-03, 7.67631886355405e-05, 2.28521601674098e-03}},
{{1.89940391362152e+02, -4.23280856852379e-03, -2.70608873541399e-02},
{-4.23280856852379e-03, 6.77547582742563e-05, 2.69154203800467e-05},
{-2.70608873541399e-02, 2.69154203800467e-05, 3.88574543373470e-03}}};
static const double kVoiceGmmMean[kVoiceGmmNumMixtures][kVoiceGmmDim] = {
{-2.15020241646536e+00, 4.97079062999877e+02, 4.77078119504505e+02},
{-8.92097680029190e-01, 5.92064964199921e+02, 1.81045145941059e+02},
{-1.29435784144398e+00, 4.98450293410611e+02, 1.71991263804064e+02},
{-1.03925228397884e+00, 4.99511274321571e+02, 1.05838336539105e+02},
{-1.29229047206129e+00, 4.15026762566707e+02, 1.12861119017125e+02},
{-7.88748114599810e-01, 4.48739336688113e+02, 1.89784216956337e+02},
{-8.77777402332642e-01, 4.86620285054533e+02, 1.13477708016491e+02},
{-2.06465957063057e+00, 6.33385049870607e+02, 2.32758546796149e+02},
{-6.98893789231685e-01, 5.93622051503385e+02, 1.92536982473203e+02},
{-2.55901217508894e+00, 1.55914919756205e+03, 1.39769980835570e+02},
{-1.92070024165837e+00, 4.87983940444185e+02, 1.02745468128289e+02},
{-7.29187507662854e-01, 5.22717685022855e+02, 1.16377942283991e+02}};
static const double kVoiceGmmWeights[kVoiceGmmNumMixtures] = {
-1.39789694361035e+01,
-1.19527720202104e+01,
-1.32396317929055e+01,
-1.09436815209238e+01,
-1.13440027478149e+01,
-1.12200721834504e+01,
-1.02537324043693e+01,
-1.60789861938302e+01,
-1.03394494048344e+01,
-1.83207938586818e+01,
-1.31186044948288e+01,
-9.52479998673554e+00};
#endif // WEBRTC_MODULES_AUDIO_PROCESSING_VAD_VOICE_GMM_TABLES_H_

View File

@ -161,8 +161,15 @@
'audio_processing/aec/system_delay_unittest.cc',
# TODO(ajm): Fix to match new interface.
# 'audio_processing/agc/agc_unittest.cc',
'audio_processing/agc/agc_audio_proc_unittest.cc',
'audio_processing/agc/circular_buffer_unittest.cc',
'audio_processing/agc/gmm_unittest.cc',
'audio_processing/agc/histogram_unittest.cc',
'audio_processing/agc/mock_agc.h',
'audio_processing/agc/pitch_based_vad_unittest.cc',
'audio_processing/agc/pitch_internal_unittest.cc',
'audio_processing/agc/pole_zero_filter_unittest.cc',
'audio_processing/agc/standalone_vad_unittest.cc',
'audio_processing/beamformer/complex_matrix_unittest.cc',
'audio_processing/beamformer/covariance_matrix_generator_unittest.cc',
'audio_processing/beamformer/matrix_unittest.cc',
@ -180,14 +187,6 @@
'audio_processing/transient/wpd_node_unittest.cc',
'audio_processing/transient/wpd_tree_unittest.cc',
'audio_processing/utility/delay_estimator_unittest.cc',
'audio_processing/vad/gmm_unittest.cc',
'audio_processing/vad/pitch_based_vad_unittest.cc',
'audio_processing/vad/pitch_internal_unittest.cc',
'audio_processing/vad/pole_zero_filter_unittest.cc',
'audio_processing/vad/standalone_vad_unittest.cc',
'audio_processing/vad/vad_audio_proc_unittest.cc',
'audio_processing/vad/vad_circular_buffer_unittest.cc',
'audio_processing/vad/voice_activity_detector_unittest.cc',
'bitrate_controller/bitrate_allocator_unittest.cc',
'bitrate_controller/bitrate_controller_unittest.cc',
'bitrate_controller/send_side_bandwidth_estimation_unittest.cc',

View File

@ -18,12 +18,12 @@
#include "gflags/gflags.h"
#include "testing/gtest/include/gtest/gtest.h"
#include "webrtc/modules/audio_processing/agc/agc.h"
#include "webrtc/modules/audio_processing/agc/agc_audio_proc.h"
#include "webrtc/modules/audio_processing/agc/common.h"
#include "webrtc/modules/audio_processing/agc/histogram.h"
#include "webrtc/modules/audio_processing/agc/pitch_based_vad.h"
#include "webrtc/modules/audio_processing/agc/standalone_vad.h"
#include "webrtc/modules/audio_processing/agc/utility.h"
#include "webrtc/modules/audio_processing/vad/vad_audio_proc.h"
#include "webrtc/modules/audio_processing/vad/common.h"
#include "webrtc/modules/audio_processing/vad/pitch_based_vad.h"
#include "webrtc/modules/audio_processing/vad/standalone_vad.h"
#include "webrtc/modules/interface/module_common_types.h"
static const int kAgcAnalWindowSamples = 100;
@ -75,7 +75,7 @@ class AgcStat {
: video_index_(0),
activity_threshold_(kDefaultActivityThreshold),
audio_content_(Histogram::Create(kAgcAnalWindowSamples)),
audio_processing_(new VadAudioProc()),
audio_processing_(new AgcAudioProc()),
vad_(new PitchBasedVad()),
standalone_vad_(StandaloneVad::Create()),
audio_content_fid_(NULL) {
@ -155,7 +155,7 @@ class AgcStat {
double activity_threshold_;
double video_vad_[kMaxNumFrames];
rtc::scoped_ptr<Histogram> audio_content_;
rtc::scoped_ptr<VadAudioProc> audio_processing_;
rtc::scoped_ptr<AgcAudioProc> audio_processing_;
rtc::scoped_ptr<PitchBasedVad> vad_;
rtc::scoped_ptr<StandaloneVad> standalone_vad_;