1805270 - Debian Testing, Macbook Pro: Degraded webrtc audio quality (~choppy,metallic,under a pillow) with both USB and integrated mic since 2021-11-01 (possibly bug 1654112?)

Status: RESOLVED WORKSFORME

(Core :: Audio/Video, defect, P3)

Description

[Darkspirit]

Attachment: mozregression_2022-12-12_aboutsupport.txt

MOZ_DISABLE_CONTENT_SANDBOX=1 mozregression --good 2021-01-01 --bad 2022-12-12 --pref javascript.options.asmjs:false javascript.options.baselinejit:false javascript.options.blinterp:false javascript.options.ion:false -a https://webrtc.github.io/samples/src/content/getusermedia/audio/

29:33.82 INFO: Last good revision: 4f57142ccd08255c6e87390db6772db94a1033c4 (2021-10-31)
29:33.82 INFO: First bad revision: 08eb1047d841ce4f1fcee1595dc4ec7ed8cba8f5 (2021-11-01)
29:33.82 INFO: Pushlog:
https://hg.mozilla.org/mozilla-central/pushloghtml?fromchange=4f57142ccd08255c6e87390db6772db94a1033c4&tochange=08eb1047d841ce4f1fcee1595dc4ec7ed8cba8f5

Comment 1

[BugBot [:suhaib / :marco/ :calixte]]

:pehrsons, since you are the author of the regressor, bug 1654112, could you take a look? Also, could you set the severity field?

For more information, please visit auto_nag documentation.

Comment 2

[Darkspirit]

Attachment: 2021-30-31_lastgood.mp4

MOZ_DISABLE_CONTENT_SANDBOX=1 mozregression --launch 2021-10-31 --pref javascript.options.asmjs:false javascript.options.baselinejit:false javascript.options.blinterp:false javascript.options.ion:false -a https://webrtc.github.io/samples/src/content/getusermedia/audio/

Comment 3

[Darkspirit]

Attachment: 2021-11-01_firstbad.mp4

MOZ_DISABLE_CONTENT_SANDBOX=1 mozregression --launch 2021-11-01 --pref javascript.options.asmjs:false javascript.options.baselinejit:false javascript.options.blinterp:false javascript.options.ion:false -a https://webrtc.github.io/samples/src/content/getusermedia/audio/

Comment 4

[Darkspirit]

Attachment: 2022-12-12_bad.mp4

MOZ_DISABLE_CONTENT_SANDBOX=1 mozregression --launch 2022-12-12 --pref javascript.options.asmjs:false javascript.options.baselinejit:false javascript.options.blinterp:false javascript.options.ion:false -a https://webrtc.github.io/samples/src/content/getusermedia/audio/

Comment 5

[Darkspirit]

Additionally, when I don't talk in Discord and don't have a green circle, friends sometimes hear noise from me which can be stopped by saying something. Wild and likely wrong guess: It's as if Firefox heard me breathing once and then some noise suppression goes crazy, I don't know.

Comment 6

[Darkspirit]

Attachment: mozregression_2022-12-12_aboutsupport_raw.json

Comment 7

[Darkspirit]

Sometimes audio becomes as bad again as what you hear in the beginning. Sometimes audio stays this bad. But that's not reliably reproducible.
What I've recorded is what's easily reproducible for me.

Comment 8

[BugBot [:suhaib / :marco/ :calixte]]

Set release status flags based on info from the regressing bug 1654112

Comment 9

[Andreas Pehrson [:pehrsons]]

(In reply to Darkspirit from comment #4)

Created attachment 9307893 [details]
2022-12-12_bad.mp4

MOZ_DISABLE_CONTENT_SANDBOX=1 mozregression --launch 2022-12-12 --pref javascript.options.asmjs:false javascript.options.baselinejit:false javascript.options.blinterp:false javascript.options.ion:false -a https://webrtc.github.io/samples/src/content/getusermedia/audio/

This sounds like the AEC could be confused because we tell it there will be echo (that page will play it straight back to speakers) but then, assuming you have headphones, there is none. All while the latency is very low so the input will look quite similar to the echo we told it it might hear but it didn't.

If you mute the audio element, like the situation would be on any conferencing site, your audio should sound fine. But I guess you might not be able to record it with OBS. S4 because it's not something that would manifest in a real-world application.

An audio sample of what is described in comment 5 would be interesting.

Comment 11

[Andreas Pehrson [:pehrsons]]

Note this is linux so cannot be bug 1670633. I'd say it's attributable to libwebrtc's audio processing which had a big update with bug 1654112. There's not too much we can do to improve that, but there should be continuous improvement since we update that code every release. Changing to WFM since it seems to work as intended.