All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=599575

Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?keys=agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICg8MWctQoM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICgiLT-uwoM,agxzfmNocm9tZXBlcmZyFAsSB0Fub21hbHkYgICg8PnNugkM


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

chromium-webrtc-rel-mac

***Note that although the summary appears similar to  bug 599574 , this covers a different CL range***

CL range:
http://test-results.appspot.com/revision_range?start=383461&end=383999

There is a WebRTC roll in that range:
https://codereview.chromium.org/1839343002

WebRTC changes in that roll:
https://chromium.googlesource.com/external/webrtc/trunk/webrtc.git/+log/ce55976..80eacc6

There appear to be quite a few audio related CLs in that roll. tlegrand@, can you take a look?

I think I know what this is, that machine was upgraded from 10.9 to 10.10 at March 30, which fits with the regression. I'll leave it to tlegrand@ to close the bug though (in case there may be something else to look at).

With more data, this actually looks like a real regression.

Henrik, can you take a look?

I think this is the machine upgrade as mentioned in #3.

The ChromiumWebRTC bots indicate a regression range in Chrome as https://chromium.googlesource.com/chromium/src/+log/2d2989c3627609fe10338d462b9b7950a277de1b..460980d0ce738731d81176a0635576766c3a2180 (corresponds to commit positions 383899:383999)
That range contains one WebRTC roll: https://chromium.googlesource.com/chromium/src/+/d867bb3a003a380406c175a72b8353361c1abfc7
The WebRTC roll updates from WebRTC revision 12143 to 12158.

Now, the ChromiumWebRTCFYI bots indicate a regression range in Chrome to be https://chromium.googlesource.com/chromium/src/+log/44af66b207f1b62d0c25650df728239509ce4124%5E..1ed9a4dfd34de7d19c6eb8d0bc20dac44f7d355e?pretty=fuller, which corresponds to commit positions 383742:383783.
That range contains another WebRTC roll (12119:12143), but since this is FYI-bots, we need to look at the WebRTC revision that the bot imports itself. Looking at where the graph was definitely "good", and contrasting it with the start of the regression, I get a WebRTC blame range of 12128:12145 (the start of the range is probably too early). The overlap with the above WebRTC range is only commits 12144 and 12145, which both look unrelated to me.

Summing up, the WebRTC and WebRTCFYI bots fired at the same time, but for different Chrome ranges, and with WebRTC ranges that overlap only by two seemingly unrelated changes.