tools/metrics/histograms/histograms.xml

Index: tools/metrics/histograms/histograms.xml
diff --git a/tools/metrics/histograms/histograms.xml b/tools/metrics/histograms/histograms.xml
index c93a3e4fde61ad846cfec8685dcc8094dd64d4bc..d47dba1b0a464737793d8caa20623343f2e6a53f 100644
--- a/tools/metrics/histograms/histograms.xml
+++ b/tools/metrics/histograms/histograms.xml
@@ -62683,7 +62683,7 @@ http://cs/file:chrome/histograms.xml - but prefer this file for new entries.
 </histogram>
 
 <histogram name="WebRTC.AecDelayBasedQuality" enum="DelayBasedEchoQuality">
-  <owner>bjornv@chromium.org</owner>
+  <owner>hlundin@chromium.org</owner>
   <summary>
     Captures if the estimated delay between rendered and captured audio is out
     of bounds which can cause the Echo Cancellation to fail. This is logged
@@ -62693,6 +62693,18 @@ http://cs/file:chrome/histograms.xml - but prefer this file for new entries.
   </summary>
 </histogram>
 
+<histogram name="WebRTC.AecDivergentFilterNonZeroPeriods" units="%">

[minyue, 2016/06/28 08:18:41]

"Periods" may be confusing

The full name should be
AecDivergentFilterFractionNonZeroPercentage
It is obviously too complicated. 

Alternative 2: AecDivergentFilterPercentage
This is too similar to AecDivergentFilterFraction and can make people think they
are the same

Alternative 3: AecFilterHasDivergencePercentage
Emm, I think this is nicer

[Henrik Grunell, 2016/06/28 09:33:10]

I like (3), but I think it can be slimmed down to AecFilterDivergence. The unit
("Percentage") can be skipped in the name. It's seen in the histogram, thus
redundant. Wdyt? Changed.

[minyue, 2016/06/28 10:01:26]

Yes, it is good to remove percentage. But I am not very much in favor of
removing "Has", since FilterDivergence% means how much time it diverged, and
with "Has", it means whether it is showed divergence. The latter implies that we
do not continuously check divergence, and we only check "whether it had
divergence in past 1 second".

[Henrik Grunell, 2016/06/28 10:09:28]

That makes sense. Done.

+  <owner>grunell@chromium.org</owner>
+  <owner>minyue@chromium.org</owner>
+  <summary>
+    The percentage of measurement periods during a capture stream's lifetime
+    when the echo canceler's filter is considered to have diverged at least
+    once. A diverged filter could mean that there was echo experienced. The

[minyue, 2016/06/28 08:18:41]

echo -> echo or ducking

[Henrik Grunell, 2016/06/28 09:33:10]

Done.

+    measurement period size is in the order of one second. Capture streams with
+    shorter lifetimes (i.e. no data available) is not recorded here.

[minyue, 2016/06/28 08:18:42]

Have you checked data comes in right after 1 second? Which test can I run to
test it?

[Henrik Grunell, 2016/06/28 09:33:10]

It doesn't come in until after >2 seconds, then it seems yo be updated every 1
second.

You can use this CL and add
CHECK_GE(metrics.divergent_filter_fraction, 0.0f);
at line 371 in media_stream_audio_processor_options.cc. It will crash, and
you'll need to increase the check period from 1 second to over 2 seconds to
avoid it.

I added a printf in the update of the fraction in AEC, and saw it was updated
with an interval that seemed to be 1 second.

+  </summary>
+</histogram>
+
 <histogram name="WebRTC.ApplicationMaxConsecutiveBytesDiscard">
   <obsolete>
     Deprecated as of 12/2014, replaced by