Pull the Voice Activity Detector out from the AGC

webrtc/modules/audio_processing/agc/agc.cc

 /*
  *  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.
  */
 
 #include "webrtc/modules/audio_processing/agc/agc.h"
 
 #include <cmath>
 #include <cstdlib>
 
 #include <algorithm>
+#include <vector>
 
-#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"
 
 namespace webrtc {
 namespace {
 
 const int kDefaultLevelDbfs = -18;
-const double kDefaultVoiceValue = 1.0;
 const int kNumAnalysisFrames = 100;
 const double kActivityThreshold = 0.3;
 
 }  // namespace
 
 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()),
-      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)) {
+      inactive_histogram_(Histogram::Create()) {
   }
 
 Agc::~Agc() {}
 
 float Agc::AnalyzePreproc(const int16_t* audio, int length) {
   assert(length > 0);
   int num_clipped = 0;
   for (int i = 0; i < length; ++i) {
     if (audio[i] == 32767 || audio[i] == -32768)
       ++num_clipped;
   }
   return 1.0f * num_clipped / length;
 }
 
 int Agc::Process(const int16_t* audio, int length, int sample_rate_hz) {
-  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];
-    }
+  vad_.ProcessCaptureAudio(audio, length);
+  std::vector<double> rms = vad_.chunkwise_rms();
+  std::vector<double> p = vad_.chunkwise_voice_probabilities();

[Andrew MacDonald, 2015/06/15 04:32:25]

Could use const auto for these. Perhaps probabilities instead of just p?

[aluebs-webrtc, 2015/06/16 01:17:52]

I find the explicit types add readability here, but I totally agree that
"probabilities" is a better naming.

+  for (size_t i = 0; i < rms.size(); ++i) {

[Andrew MacDonald, 2015/06/15 04:32:25]

DCHECK_EQ(rms.size(), p.size()) ?

[aluebs-webrtc, 2015/06/16 01:17:52]

Done.

+    histogram_->Update(rms[i], p[i]);
   }
   return 0;
 }
 
 bool Agc::GetRmsErrorDb(int* error) {
   if (!error) {
     assert(false);
     return false;
   }
 
@@ skipping 21 matching lines @@
   // TODO(turajs): just some arbitrary sanity check. We can come up with better
   // limits. The upper limit should be chosen such that the risk of clipping is
   // low. The lower limit should not result in a too quiet signal.
   if (level >= 0 || level <= -100)
     return -1;
   target_level_dbfs_ = level;
   target_level_loudness_ = Dbfs2Loudness(level);
   return 0;
 }
 
-void Agc::EnableStandaloneVad(bool enable) {
-  standalone_vad_enabled_ = enable;
-}
-
 }  // namespace webrtc