1263248 - Echo cancellation doesn't work well anymore

Status: RESOLVED FIXED

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

Description

[Marco Castelluccio [:marco]]

It is a recent regression. I'm not seeing the same behavior with other VoIP software (e.g. Vidyo).

I'm seeing this on https://apprtc.appspot.com/.

Comment 1

[Randell Jesup [:jesup] (needinfo me)]

Just landed a patch which should have fixed that.  On inbound; you can pull it and build or wait for merge and a nightly.  Please retest with that.  Thanks!

What OS and FF version, by the way?  Note bug 1263251 affects all OSes.

Comment 2

[Marco Castelluccio [:marco]]

(In reply to Randell Jesup [:jesup] from comment #1)
> What OS and FF version, by the way?  Note bug 1263251 affects all OSes.

Firefox 48 on Linux 64-bit (Ubuntu).
I see bug 1250934 landed on 2016-03-08. I'm sure I've used WebRTC *after* 2016-03-08 and I've noticed the regression just a few days ago, so there's a possibility this is another bug (or maybe the odd race condition was always hit for me and stopped being hit a few days ago).

Comment 3

[Randell Jesup [:jesup] (needinfo me)]

Note: if you hear echo, it's coming from the *other* person's machine, not yours.  So what they're running is what matters.  If people hear their echo from you, then yours matters.

Now we also preffed on full-duplex in Linux not too long ago (after the bug 1250934 landing), but the error should have broken all AEC use, full_duplex or not, after bug 1250934.  Note that if the person is using a headset, the AEC being busted generally won't be noticed.

(BTW, I landed the fix 11 minutes after you reported this bug - and hours before I saw the bug ;-)

I'm very interested in if this helps with your problem - and remember my first point about who will hear the echo, and who has to update!  I've already filed an uplift request for 47.  Thanks!

Comment 4

[Marco Castelluccio [:marco]]

I'll add more details:
- The other person is using Firefox Beta on Windows 10;
- We were both hearing echo, but it was so noisy that it was hard to tell what it was;
- We stop hearing echo when I use headphones.

Comment 5

[Marco Castelluccio [:marco]]

Much better now, there's still some weird echo from time to time. I'll try to record it next time.

Should I file a new bug?

Comment 6

[Randell Jesup [:jesup] (needinfo me)]

Sure - and if you can, when you hear the echo, have the person *not* hearing the echo switch to an about:webrtc tab and hit "Start AEC Logging", chat for a minute, and then Stop AEC Logging (it will stop on it's own after a few minutes, but the button won't show that).  Attach all the aec files in the directory indicated (zip/tar them up).  Include if either person was on a headset and what version each was running.  Thanks!

Comment 7

[Marco Castelluccio [:marco]]

(In reply to Randell Jesup [:jesup] from comment #6)
> Sure - and if you can, when you hear the echo, have the person *not* hearing
> the echo switch to an about:webrtc tab and hit "Start AEC Logging", chat for
> a minute, and then Stop AEC Logging (it will stop on it's own after a few
> minutes, but the button won't show that).  Attach all the aec files in the
> directory indicated (zip/tar them up).  Include if either person was on a
> headset and what version each was running.  Thanks!

We both hear the echo. I will collect the AEC log anyway.

Comment 8

[Marco Castelluccio [:marco]]

I haven't been able to reproduce it anymore.

Comment 9

[Randell Jesup [:jesup] (needinfo me)]

Good - thanks!  Also, additional AEC fixes have just landed (as of 4/20 nightly), and one or two others recently.

Comment 10

[Marco Castelluccio [:marco]]

Attachment: My Recording #4.mp4

I've been able to reproduce it again.
This is the echo that I hear.

I've asked the other person to use headphones, I don't hear the echo anymore but she does (although different, as she hears her voice clearly and not that weird sound).

Comment 11

[Randell Jesup [:jesup] (needinfo me)]

Note: AEC is turned off after the first use of getUserMedia in the 5/25 (and perhaps 5/24) Nightlies, so if you're using that, please ignore until bug 1275703 lands (probably today; hopefully will merge for tomorrow's Nightly)

Comment 12

[Marco Castelluccio [:marco]]

I've been hearing intermittently in the last few weeks, so it should be unrelated.

Comment 13

[Maire Reavy [:mreavy]]

Hi Marco -- Regarding the echo you are hearing now:  Can you describe it to us?  Are you still using Firefox 48 on Linux?  Are you hearing the echo or is your calling partner hearing it or are you both hearing echo?  (As a quick reminder, which you probably know already but it's easy to forget: if you are hearing the echo, the problem/echo is being caused by the device on the other side of the call, not your machine/Firefox.)

Would you be available for any test calls this week?  What time zone are you in this week?  Thanks.

Comment 14

[Marco Castelluccio [:marco]]

(In reply to Maire Reavy [:mreavy] (Plz ni?:mreavy) from comment #13)
> Hi Marco -- Regarding the echo you are hearing now:  Can you describe it to
> us?

I've attached a recording, can you hear it?

> Are you still using Firefox 48 on Linux?

I'm using Nightly (currently 50) on Linux (Ubuntu 16.04), my partner is using Firefox Beta (I guess 47) on Windows 10.

> Are you hearing the echo or is your calling partner hearing it or are you both hearing echo?

We're both hearing echo. If either one of us uses headphones, the person who doesn't use headphones
doesn't hear the echo anymore, the person who uses them does hear the echo (although a "clear" echo, different
than the one you can hear in the recording).

> Would you be available for any test calls this week?  What time zone are you
> in this week?  Thanks.

Yes, I'm in London.

Comment 15

[David Patten]

Hello. We are experiencing the exact issue as reported by Marco. When clients use our application with Firefox, our Agents who are using Chrome, all experience echo even when they are using headsets. Without changing audio devices, having the client use Chrome eliminates the echo heard by the Agent. We are promoting Firefox in our business model and currently in production, however, the echo problem is highly impacting our ability to serve customers when our Agents are experiencing the echo. We are open to engaging into any type of testing to help isolate the problem. My dev team is located in Madrid, Spain and our corporate location is in Florida. Any assistance would be greatly appreciated.

Comment 16

[Maire Reavy [:mreavy]]

David -- Thanks for chiming in on this bug.  I'm just back from PTO.  (Marco -- thanks for needinfo'ing me.) Can you tell me what versions of Firefox and Chrome you are using?  

Marco -- Does the echo appear at the start, or later?  Does it go away after a while (a few to 30 seconds)?  Can you please retest?  The recording you uploaded appears to be from a period of time when the AEC was broken (no cancellation) in Nightly in June.  Also, can you do test calls with us?  We may have a test build for you to try with a fix to an upstream AEC bug.

David -- Would you be willing to do test calls with me and/or Jesup (the bug owner)?  

We've made a bunch of changes to the audio to improve quality and more are coming, and I think you found some regressions that we have since fixed (fingers crossed).

Thanks!

Comment 17

[Marco Castelluccio [:marco]]

I haven't noticed this bug in a while, so I would close it as WORKSFORME.

David, have you noticed it?

Comment 18

[Paul Adenot (:padenot)]

That's kind of expected, we've made tons of changes to the audio stack recently, thanks for the feedback.

Comment 19

[Bulk Bug Changes for mreavy's org]

Mass change P1->P2 to align with new Mozilla triage process

Comment 20

[Marco Castelluccio [:marco]]

I can reproduce it again. Firefox (Nightly and Beta) on Windows on both ends.

Comment 21

[Marco Castelluccio [:marco]]

(In reply to Marco Castelluccio [:marco] from comment #20)
> I can reproduce it again. Firefox (Nightly and Beta) on Windows on both ends.

To be precise, this must be a new bug. I can't reproduce it on 55 but I can on 57.
A quick mozregression run pointed me to https://hg.mozilla.org/integration/mozilla-inbound/json-pushes?changeset=5bbdb7d36ee3c136a0ed03be9d5b012d05dfd08e&full=1.

I will open a new bug since it's likely a different cause.

Comment 22

[Sylvestre Ledru [:Sylvestre]]

Moving to p3 because no activity for at least 1 year(s).
See https://github.com/mozilla/bug-handling/blob/master/policy/triage-bugzilla.md#how-do-you-triage for more information

Comment 23

[Sylvestre Ledru [:Sylvestre]]

Moving to p3 because no activity for at least 1 year(s).
See https://github.com/mozilla/bug-handling/blob/master/policy/triage-bugzilla.md#how-do-you-triage for more information

Comment 24

[Dan Minor [:dminor]]

Based on comment 21, this bug was fixed as filed (as was the bug for the new issue). Closing and clearing old needinfo.