Adding fraction of filter divergence in AEC metrics.

webrtc/modules/audio_processing/aec/aec_core.cc

Index: webrtc/modules/audio_processing/aec/aec_core.cc
diff --git a/webrtc/modules/audio_processing/aec/aec_core.cc b/webrtc/modules/audio_processing/aec/aec_core.cc
index 2cc2968f81c2e2ffc658d15ba048e3890d8a4a47..4f07a00800c756dc40458f518a13e86741ca75fa 100644
--- a/webrtc/modules/audio_processing/aec/aec_core.cc
+++ b/webrtc/modules/audio_processing/aec/aec_core.cc
@@ -48,6 +48,11 @@ static const int subCountLen = 4;
 static const int countLen = 50;
 static const int kDelayMetricsAggregationWindow = 1250;  // 5 seconds at 16 kHz.
 

[tlegrand-webrtc, 2016/03/02 15:30:29]

I'm wondering if "Diverge" is the right term, or if we should call it
"Divergence". What do you guys think makes most sense?

[kwiberg-webrtc, 2016/03/02 18:33:03]

I believe "Divergent" is the trendy form.

+// Diverge metric is based on audio level, which gets updated which every

[tlegrand-webrtc, 2016/03/02 15:30:29]

remove "which" in "which every"

+// |subCountLen + 1| * 10 milliseconds. Diverge metric takes the statistics of
+// |kDivergeMetricAggregationWindow| samples.
+static const int kDivergeMetricAggregationWindow = 50;  // 1 seconds at 16 kHz.
+
 // Quantities to control H band scaling for SWB input
 static const float cnScaleHband = 0.4f;  // scale for comfort noise in H band.
 // Initial bin for averaging nlp gain in low band
@@ -554,6 +559,10 @@ static void InitMetrics(AecCore* self) {
   InitStats(&self->erle);
   InitStats(&self->aNlp);
   InitStats(&self->rerl);
+
+  self->num_diverge_values = 0;
+  self->num_filter_diverge = 0;
+  self->fraction_filter_diverge = 0.0f;
 }
 
 static float CalculatePower(const float* in, size_t num_samples) {
@@ -611,6 +620,20 @@ static void UpdateMetrics(AecCore* aec) {
     aec->stateCounter++;
   }
 
+  if (aec->linoutlevel.sfrcounter == 0) {
+    // We count a filter divergence when the audio level of linear filter output
+    // is larger than that of near-end signal by -30dB or more.

[peah-webrtc, 2016/02/29 14:05:56]

I don't think the usage of "-30dB" is correct. I think that it should be 0.0043
dB.

I.e. (including some rephrasing):
// We count it as a filter divergence when the audio level of the linear filter
output
// is larger than that of the near-end signal by 0.0043 dB or more.

+    if (aec->linoutlevel.framelevel > 1.001 * aec->nearlevel.framelevel)
+      aec->num_filter_diverge++;
+    aec->num_diverge_values++;
+    if (aec->num_diverge_values >= kDivergeMetricAggregationWindow) {
+      aec->fraction_filter_diverge =
+          static_cast<float>(aec->num_filter_diverge) / aec->num_diverge_values;
+      aec->num_diverge_values = 0;
+      aec->num_filter_diverge = 0;
+    }
+  }
+
   if (aec->farlevel.frcounter == 0) {
     if (aec->farlevel.minlevel < noisyPower) {
       actThreshold = actThresholdClean;