Pull the Voice Activity Detector out from the AGC

webrtc/modules/audio_processing/agc/agc.cc

Index: webrtc/modules/audio_processing/agc/agc.cc
diff --git a/webrtc/modules/audio_processing/agc/agc.cc b/webrtc/modules/audio_processing/agc/agc.cc
index 6041435bd91ddf433b0687e366e41b9be1f452d1..54b56eeabc505138da3dfd785933d70eb4f00294 100644
--- a/webrtc/modules/audio_processing/agc/agc.cc
+++ b/webrtc/modules/audio_processing/agc/agc.cc
@@ -14,13 +14,10 @@
 #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/base/checks.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"
 
@@ -28,7 +25,6 @@ namespace webrtc {
 namespace {
 
 const int kDefaultLevelDbfs = -18;
-const double kDefaultVoiceValue = 1.0;
 const int kNumAnalysisFrames = 100;
 const double kActivityThreshold = 0.3;
 
@@ -36,16 +32,9 @@ 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()),
-      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() {}
@@ -61,55 +50,12 @@ float Agc::AnalyzePreproc(const int16_t* audio, int 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_.ProcessChunk(audio, length, sample_rate_hz);
+  std::vector<double> rms = vad_.chunkwise_rms();
+  std::vector<double> probabilities = vad_.chunkwise_voice_probabilities();

[Andrew MacDonald, 2015/06/16 22:18:31]

You're still allocating two vectors here and copy assigning the return values to
them. You want:

const std::vector<double>& rms = ...

[aluebs-webrtc, 2015/06/17 01:44:31]

You are right. Done.

+  DCHECK_EQ(rms.size(), probabilities.size());
+  for (size_t i = 0; i < rms.size(); ++i) {
+    histogram_->Update(rms[i], probabilities[i]);
   }
   return 0;
 }
@@ -151,8 +97,4 @@ int Agc::set_target_level_dbfs(int level) {
   return 0;
 }
 
-void Agc::EnableStandaloneVad(bool enable) {
-  standalone_vad_enabled_ = enable;
-}
-
 }  // namespace webrtc