All graphs for this bug:
  https://chromeperf.appspot.com/group_report?bug_id=765564

(For debugging:) Original alerts at time of bug-filing:
  https://chromeperf.appspot.com/group_report?sid=5aac0b7e57a47fa07a5afb26fe5575e368b236fa015683dbbe8e5d557fec684b


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

chromium-webrtc-rel-win10

This machine was replaced around Aug 11 as part of bug 754457 but it doesn't explain this regression, which happened on Aug 17. 

I assume this audio metric is affected by the audio device, or? I don't have anything to do here, so I'm bouncing it back to you to decide hlundin@.

But, the CL I was pointing to was about "Add Debug builders to chromium.webrtc.fyi". Can that have an impact? Or was it that you commented on in #2, but I didn't understand?

Sorry for not answering the original question. Normally the testers shouldn't possibly be affected by adding the debug builders, but in this case they were, since a bug surfaced when I added them, making them send builds to the testers. This was fixed quickly at Aug 18 though: https://chromium-review.googlesource.com/c/chromium/tools/build/+/620588
so I cannot explain why it would affect the testers all the following builds, especially as this regression only covers Win10.

If you look at the last two days, you can see that most data points starting in Oct 7 seem to actually have recovered?

Yes, it seems to be recovering. I'll monitor it a few more days before closing.

I'm declaring that this has recovered.