Tweak kDecayRate in the IntelligibilityEnhancer

webrtc/modules/audio_processing/intelligibility/intelligibility_enhancer.cc

 /*
  *  Copyright (c) 2014 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.
  */
 
@@ skipping 16 matching lines @@
 const size_t kErbResolution = 2;
 const int kWindowSizeMs = 16;
 const int kChunkSizeMs = 10;  // Size provided by APM.
 const float kClipFreqKhz = 0.2f;
 const float kKbdAlpha = 1.5f;
 const float kLambdaBot = -1.0f;      // Extreme values in bisection
 const float kLambdaTop = -1e-5f;      // search for lamda.
 const float kVoiceProbabilityThreshold = 0.02f;
 // Number of chunks after voice activity which is still considered speech.
 const size_t kSpeechOffsetDelay = 80;
-const float kDecayRate = 0.98f;              // Power estimation decay rate.
-const float kMaxRelativeGainChange = 0.04f;  // Maximum relative change in gain.
+const float kDecayRate = 0.994f;              // Power estimation decay rate.

[peah-webrtc, 2016/04/01 04:43:50]

I cannot really say that I yet understand the underlying algorithm but it seems
a bit strange to lock the map the relative_change_limit in GainApplier to the
decay rate.

I'm not saying it is wrong to do so, but they seem quite independent, and are
used in different places. Furthermore, this is a change from how it was before.

[hlundin-webrtc, 2016/04/01 08:36:38]

Acknowledged.

[aluebs-webrtc, 2016/04/01 17:33:35]

Well, the decay rate of the PSD estimations doesn't necessarily have to be
linked to the maximum gain change, but it kind of makes sense, since both are
linked to the speed the algorithm can adapt. And I find it elegant to only have
one constant to tweak the adaptation speed. But of course, I can go back to
having 2 different constants if you think it adds value.

 const float kRho = 0.0004f;  // Default production and interpretation SNR.
 
 // Returns dot product of vectors |a| and |b| with size |length|.
 float DotProduct(const float* a, const float* b, size_t length) {
   float ret = 0.f;
   for (size_t i = 0; i < length; ++i) {
     ret += a[i] * b[i];
   }
   return ret;
 }
@@ skipping 22 matching lines @@
       sample_rate_hz_(sample_rate_hz),
       num_render_channels_(num_render_channels),
       clear_power_estimator_(freqs_, kDecayRate),
       noise_power_estimator_(num_noise_bins, kDecayRate),
       filtered_clear_pow_(bank_size_, 0.f),
       filtered_noise_pow_(num_noise_bins, 0.f),
       center_freqs_(bank_size_),
       capture_filter_bank_(CreateErbBank(num_noise_bins)),
       render_filter_bank_(CreateErbBank(freqs_)),
       gains_eq_(bank_size_),
-      gain_applier_(freqs_, kMaxRelativeGainChange),
+      gain_applier_(freqs_, 1.f - kDecayRate),
       audio_s16_(chunk_length_),
       chunks_since_voice_(kSpeechOffsetDelay),
       is_speech_(false),
       noise_estimation_buffer_(num_noise_bins),
       noise_estimation_queue_(kMaxNumNoiseEstimatesToBuffer,
                               std::vector<float>(num_noise_bins),
                               RenderQueueItemVerifier<float>(num_noise_bins)) {
   RTC_DCHECK_LE(kRho, 1.f);
 
   const size_t erb_index = static_cast<size_t>(
@@ skipping 214 matching lines @@
   vad_.ProcessChunk(audio_s16_.data(), chunk_length_, sample_rate_hz_);
   if (vad_.last_voice_probability() > kVoiceProbabilityThreshold) {
     chunks_since_voice_ = 0;
   } else if (chunks_since_voice_ < kSpeechOffsetDelay) {
     ++chunks_since_voice_;
   }
   return chunks_since_voice_ < kSpeechOffsetDelay;
 }
 
 }  // namespace webrtc