All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=606681

Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICgxPuQtwoM


Bot(s) for this bug's original alert(s):

chromium-webrtc-trunk-tot-rel-mac

The alert contains no blamelist, but by traversing the graph backwards it seems as if the issue was triggered already at Point ID: 388774

@pbos: Your CL https://codereview.webrtc.org/1904983002 is listed as a possible cause for the issue arising at Point ID 388774. Could you please take a look to see whether that CL could have caused this issue?

After talking to @pbos, that CL is most likely not the cause of this.
@kjellander: Could this be due to a HW upgrade on the bot machine? A simular change was flagged as being caused by that on Mach 30-2016. 

Re #4: Nobody has touched this machine recently so it should not be changed in any way (then I would have known).

@hlundin: Do you know anything that could have caused the goog_speech_expand_rate to change? It sounds like it is related to neteq but I guess it could as easily be related to something with the network handling. 

The same change happens on the non-FYI bot three days later. But no alert has been issued (cc: kjellander for this). See https://chromeperf.appspot.com/report?sid=9244f9494d6ab6afec8c61fe32c59254a8709ec1d3963e2a848ce717bc8dc1ce&start_rev=388434&end_rev=390362.

The blame range for that regression contains a WebRTC roll and four very unrelated chrome CLs: https://chromium.googlesource.com/chromium/src/+log/278873bbaf13b728fc2be5834067367ba45297dd..ab8e57c751b135139f3666ad900f491e78744fcd

The range for the WebRTC roll contains pbos's CL in #3 (cc: pbos), and a large number of other CLs. So, it seems that something is actually going on there.

This is an improvement, since low expand rates are good, but I'd like to understand why.

cc:ivoc, next week's perf sheriff.

Any update on this bug?

Re #4: I normally try to remember to notify our perf sheriffs when there has been an upgrade, but it's easy to forget.
AFAIK, there hasn't been any upgrades/hardware changes in March 30 (which affected bugs like  bug 600508 ).

Ping, should we just close this bug?

Yes, I'll close this now. I don't know why it happened, but (1) the change is an improvement and (2) it matches an earlier regression ( issue 599574 ) that was attributed to HW upgrade -- maybe it wasn't.