User Agent: Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:81.0) Gecko/20100101 Firefox/81.0

Steps to reproduce:

I connect to google meet to join a meeting but

Actual results:

I had many problems with intermittent audio., also other colleagues report this problem (but I don't know if they used firefox). I suggest to move to google chrome and after the people that has problems move to chrome the problem are solved

Expected results:

we expect to use google meet also on other browsers

Hi Marco,

Thank you for your bug report! When you say you have problems with intermittent audio, is the problem with others hearing intermittent audio when you speak, or you hearing intermittent audio when others speak, or possibly both? Do you know if this is a new problem with Firefox 81, or have you noticed this with earlier versions? Thanks!

Hello Dan,
We are also experiencing this problem. For us it's usually that people who join the meeting get an audio delay of about ~1 - 2 seconds. This means I won't hear anything the other person says for a second and after that the audio will be slightly out of sync with the video.
Let me know if I can provide more details.

Hello,

I have the same problem, since yesterday (29/10).
I got the reconnect indication during the meeting, but I was still connected. I also had real disconnection. The other navigation pages also happen to be slowed down.
The previous weeks, I had no slowdown in navigation and discussion.

For exemple: I write a text on google translate and it takes a few seconds before being updated on the translated part.

Os : Windows 10 64b
Firefox : 82.0.2 64b

Hi Paul, I'm not sure if there is anything actionable here right now, but there are enough reports that I think this is a real problem. Do you mind having a look at prioritization?

P1 S1 until proven otherwise. Slowness of the whole browser I've seen myself, but we don't have enough resources to work on it. The rest is a regular performance issue it sounds like.

We're looking at a policy change of asking for active assignees on S1 bugs. Is there anything we can do here?

I just reached out to the Meet team.

To the reporter here on the bug: what kind of audio equipment are you using?
Is by any chance everyone using Bluetooth head sets?

(In reply to Dan Minor [:dminor] from comment #1)

Hi Marco,

Thank you for your bug report! When you say you have problems with intermittent audio, is the problem with others hearing intermittent audio when you speak, or you hearing intermittent audio when others speak, or possibly both? Do you know if this is a new problem with Firefox 81, or have you noticed this with earlier versions? Thanks!

Hello, I had at the time of the bug report the last Firefox version, the intermittent audio was when other speak. I din't use bluethoot or other equipements, just the laptop that usially I use.

This problem disappear when I moved to Google Chrome to follow the meeting.

This bug seems bad but also, sadly, unactionable. Is there more information we could get that would allow progress? Should we mark it as stalled?

Unfortunately the Meet folks were not able to help us with these reports so far.

The bug is so annoying because I use firefox daily but I am unable to join google meets because even though I have a persistent connection the audio. I can here it for like 3 sec and it stops for about 8 sec. This repeats for the entire meet.

Firefox 90 and Firefox 91 have seemed to fixed
But Firefox 92 and The Beta version 93 are affected

Please fix this bug as fast as possible I'm temporarily migrating to chromium.

Or I'll roll back to previous versions and update if the bug can be reproduced.
If not I'll be using the older version until it's fixed

Bug 1739505 may help here. TBD once that work lands.

I have this same issue on Firefox 100.0.1 on macOS 12.3.1. My device is a MacBookPro16,1. After a couple of minutes the audio starts to be intermittent (robotic, distorted, and scratchy) in both direction: I hear the scratches, and on the other end they hear me distorted. When joining the same call in Google Chrome (with the exact device and configuration) I have no issues. Let me know if there's additional info that I can provide.

This is a different bug. Can you please describe your setup in terms of audio input and output devices? A copy of the raw data of about:support will also help.

(In reply to Paul Adenot (:padenot) from comment #14)

This is a different bug. Can you please describe your setup in terms of audio input and output devices? A copy of the raw data of about:support will also help.

Feel free to point me to a different bug, or let me know if it's necessary to file a new one. In terms of audio input/output I'm using the machine built-in microphone and speakers.

(In reply to Coral Mountain from comment #15)

(In reply to Paul Adenot (:padenot) from comment #14)

This is a different bug. Can you please describe your setup in terms of audio input and output devices? A copy of the raw data of about:support will also help.

Feel free to point me to a different bug, or let me know if it's necessary to file a new one. In terms of audio input/output I'm using the machine built-in microphone and speakers.

I think this is new, especially if you're using a recent macbook and Fx 100. You can open a new bug at https://bugzilla.mozilla.org/enter_bug.cgi?product=Core&component=Audio%2FVideo%3A%20cubeb (this goes to the right component so I'm immediately notified).

I'll provide you instructions for recording an example of audio and getting us logs so we can investigate / fix it. Thanks!

For anyone running into audio issues, please post your about support text. Also, if you can please install the firefox profiler extension and capture a 'media' profile when you experience the issue, upload the profile and post the link here. Thanks!

https://profiler.firefox.com/

From: https://github.com/webcompat/web-bugs/issues/107298#issuecomment-1185791486

I was able to catch it in a profile, here: https://share.firefox.dev/3IEKdBP. The entire time period of this profile was while the call was choppy and my CPU was downclocked.

It seems like it happens more when screen sharing is being used (another person sharing their screen with em), but not exclusively.

Replying here instead of the Github thread so I can more easily attach a file.

I've also taken a profile with the Media preset (the last one was the Firefox preset). https://share.firefox.dev/3octUTh

I agree that this looks like thermal throttling, though I don't think my laptop is getting particularly hot. It's warm to the touch, but not unusually so, and watching sensor temperatures doesn't show CPU temps above 70C, and typically around 60C. That seems like reasonable temperature to me that should not cause throttling.

I've attached my about:support data.

Yes, your machine is completely overloaded. Even the real-time audio threads don't have their scheduling slices. You probably want to vaccum it or something, that can't hut. Please also make sure the audio threads are promoted appropriately, I'm not sure one has to do something specific with Arch. To check this, do a WebRTC call (just yourself in a Meet room is enough), and run:

watch -n1 "ps -eo pid,tid,class,rtprio,ni,pri,psr,pcpu,stat,wchan:14,comm -T | grep ' RR '"

check that there are threads called AudioIPC0 with the PID of the tab. There should also be rtkit-daemon and then some PulseAudio or PipeWire threads.

Additionally, and more importantly, your machine is using WebRender with software fallback, so it's not really using the GPU to composite a bunch of videos (webcam, screen capture, that might be quite big), in addition to decoding them, and encoding yours (plus audio encoding/decoding, plus audio processing, etc.). This is the thing you want to fix first.

watch -n1 "ps -eo pid,tid,class,rtprio,ni,pri,psr,pcpu,stat,wchan:14,comm -T | grep ' RR '"

    853     964 RR      20   -  60   0  0.1 Ssl  do_epoll_wait  pipewire
    855     960 RR      20   -  60   1  0.0 Ssl  do_epoll_wait  wireplumber
    856     963 RR      20   -  60   6  0.0 SLsl do_epoll_wait  pipewire-pulse
    860     947 RR      99   - 139   7  0.0 SNsl -              rtkit-daemon
   6073    6219 RR      10   -  50   5  0.0 Sl   do_epoll_wait  AudioIP~ack RPC
   6073    6288 RR      10   -  50   1  0.0 Sl   futex_wait_que AudioIPC0
 322192  322483 RR      10   -  50   2  0.7 Sl   futex_wait_que AudioIPC0

That last row has the PID of the Google Meet tab. I'm not sure how to tell if it's being properly promoted. Does this look wrong?

Yes, your machine is completely overloaded.

I mentioned this in the Webcompat issue, but I haven't here yet. The symptom I'm observing is that while I'm on a Google Meet my CPu will drop from it's normal frequency down to it's minimum frequency (about 600MHz if memory serves) and hold there for ~10 seconds before recovering. It's during that time that the audio is choppy. In fact my entire system is pretty much unusable for those ten seconds. It's not at all surprising that the audio threads are being overloaded. This does not seem to correlate with increased CPU temperature.

Notably, this does not happen when I use another WebRTC site, https://gather.town, in Firefox and it doesn't happen with Meet in Chrome. I think the issue here is that something is causing my CPU to downclock, not that the audio gets choppy when they are starved for CPU time.

Additionally, and more importantly, your machine is using WebRender with software fallback, so it's not really using the GPU to composite a bunch of videos (webcam, screen capture, that might be quite big), in addition to decoding them, and encoding yours (plus audio encoding/decoding, plus audio processing, etc.). This is the thing you want to fix first.

Is there any documentation for how I can fix this? I didn't see anything on SuMo. I switched over to using Wayland directly instead of XWayland, and although that helped things be generally visually smoother, it didn't help with the video call issues.

That last row has the PID of the Google Meet tab. I'm not sure how to tell if it's being properly promoted. Does this look wrong?

Yes, it works well if it appears at all because of the grep RR that filters by thread that have been promoted successfuly.

Is there any documentation for how I can fix this? I didn't see anything on SuMo. I switched over to using Wayland directly instead of XWayland, and although that helped things be generally visually smoother, it didn't help with the video call issues.

No idea, #gfx-firefox on chatmo will know more, or maybe :jrmuizel knows who will know or what to do?

(In reply to Michael Cooper [:mythmon] from comment #21)

Additionally, and more importantly, your machine is using WebRender with software fallback, so it's not really using the GPU to composite a bunch of videos (webcam, screen capture, that might be quite big), in addition to decoding them, and encoding yours (plus audio encoding/decoding, plus audio processing, etc.). This is the thing you want to fix first.

Is there any documentation for how I can fix this? I didn't see anything on SuMo. I switched over to using Wayland directly instead of XWayland, and although that helped things be generally visually smoother, it didn't help with the video call issues.

You can try enabling it via prefs, visit about:config, and check the values for
gfx.webrender.enabled
gfx.webrender.fallback.software

I think I was able to get Webrender into hardware mode instead of software. The problem occured again today with that configuration. Here's about:config and a profile.

That looks better from an real-time audio point of view.

The audio threads now has a reasonnable load (about 7%). Do you still experience the same issue, where sometimes the audio goes bad for some time, and then it's better?

This still happens regularly yes. While on Google Meet calls my computer becomes unusable for ~10ish seconds at a time. As well as the audio problems, other programs slow down, my mouse lags, and sometimes I get application not responding notices from Firefox.

I noticed that when it's happening sometimes a process called "RDD" is using a lot of CPU, and also sometimes WirePlumber is using a lot of CPU. This may simple be that with the reduced pie of my downclocked CPU, any activity will pin the CPU, but maybe something weird is going on?

Although the temperatures of any particular component aren't getting too high, it does feel like a thermal limiting problem. Maybe one particular component of my processor is getting overwhelmed?

Yes, this is likely to be a thermals problem, or something else outside of our control. Maybe the computer has lots of dust in the fan intake/exhaust.

We're making optimizations to the WebRTC stack at the moment to lower the resource usage, which will mitigate the problem, but it's a real problem anyway.

I have the same issue. Firefox 102 on Kubuntu 22 LTS. No audio on the Google Meet speaker test, and no audio from other meeting attendees. Everything works fine in Chrome.

I have the "robot voice" issue with Firefox only and an external Logitech webcam (below), but using Chrome there is no problem:

Name 	Group 	Vendor 	State 	Preferred 	Format 	Channels 	Rate 	Latency
MacBook Pro Microphone	builtin-internal-mic|spk	Apple Inc.	Enabled	None	default: F32LE, support: S16LE S16BE F32LE F32BE	1	default: 48000, support: 44100 - 96000	2454 - 6535
C270 HD WEBCAM	C270 HD WEBCAM:046D:0825	Unknown Manufacturer	Enabled	All	default: F32LE, support: S16LE S16BE F32LE F32BE	1	default: 48000, support: 16000 - 48000	63 - 4144

Yes, this is likely to be a thermals problem, or something else outside of our control. Maybe the computer has lots of dust in the fan intake/exhaust.

I'm quite paranoid about resource usage (I've started programming with 1 MHz 8 bit CPUs and 5 Kb RAM...) and heating efficiency (we had a constant of > 35 C here this summer).

Therefore I'm confident in saying that CPU usage goes up at sensibly higher levels with Firefox in my case, while with Chrome it stays low.

I'm currently on an Apple Powerbook 16'' (Intel) in clamshell (closed-lid) mode.

The problem literally never happened before with open-lid/internal webcam/mic.

Hey Diego,

Could you help us out by capturing and uploading a Firefox performance profile when this happens? Visit https://profiler.firefox.com/ for more information. When capturing, please select the Media profile in the drop down.

Here you go Jim: https://share.firefox.dev/3feguVo

By the way I'm using Firefox 106b2 because I read it brings a lot of improvements about WebRTC, I tried it yesterday in a real-world scenario but I've got the exact same robotic-voice issue as with 105.

Hey diego, any chance you can do another media profile for us?

(In reply to Jim Mathies [:jimm] from comment #32)

Hey diego, any chance you can do another media profile for us?

Hey Jim, sorry for the long wait but I haven't had access to my hardware for the past month.

Also, using Firefox 117, I just did a fairly long Google Meet session (> 1hr) with almost 10 people (with screen sharing, too) using my external camera and I haven't had the issue at all.

Actually I haven't had any issue, including heating for too much CPU usage (which is higher than Chrome but still - also wondering how Chrome manages to reach such low levels of CPU usage...).

From now on, I'll restart using Firefox for my meetings and I will report as soon as anything weird happens, however my gut feeling is that this bug has been solved.

Thanks for the report back!