985714 - Report AEC echo levels (ERLE/etc)

Status: RESOLVED FIXED

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

Description

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

Attachment: aec_erle.patch

We should report echo levels in the AEC.  Note that calculating them can cost CPU; so we don't want to blindly enable this all the time.  We need to consider what the options are.  Also, it can take a while to get results after enabling.

Right now, I've made it depend on NSPR_LOG_MODULES=mediamanager:5 -- we likely should change this.  Currently the patch I have dumps the stats once every few second or so.

We should add it to about:webrtc (note: this will be complicated by the cost of measuring it/enabling/etc)

Comment 1

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

I did some profiling, and on a debug build it used <0.1% of the CPU used in an audio-only call on linux ... i.e. very little, so we can likely leave it on.

On the downside: it is *very* hard to interpret ERLE in a vacuum; it depends heavily on the input signals, noise, volume, etc.  So telemetry on ERLE may be very hard to interpret/misleading.  Also, if used with telemetry we'd want to segregate it so headphone and speakerphone uses don't get mixed (we could segregate roughly based on ERL as a stand-in for knowing if it's a headset or not).

Jean-Marc - do you know any good automated metrics beyond ERL/ERLE we could use?  I'll also consult my Voice Quality for mobile book by Perry.

Comment 2

[Jean-Marc Valin]

I think ERLE is probably a good metric here. OTOH, ERL is completely meaningless. The problem is that turning the volume up by X dB on your external speakers while turning it down X dB in the mixer will increase the ERL by X dB even though nothing actually changed in practice. The only metric I could think of would be a signal-to-echo ratio (SER). Since the signal power is fixed, then the ERLE also happens to be the improvement in SER.

Comment 3

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

Attachment: Monitor AEC echo levels (ERLE/etc) in gUM

Comment 4

[Jan-Ivar Bruaroey [:jib] (needinfo? me)]

Comment on attachment 8394951 [details] [diff] [review]
Monitor AEC echo levels (ERLE/etc) in gUM

Review of attachment 8394951 [details] [diff] [review]:
-----------------------------------------------------------------

lgtm, with the caveat that I haven't seen the patch that this patch depends on (there's no update_echo or mStarted in the tree), so bear in mind when landing.

::: content/media/webrtc/MediaEngineWebRTCAudio.cpp
@@ +589,5 @@
>  
> +#ifdef PR_LOGGING
> +  mSamples += length;
> +  if (mSamples > samplingFreq) {
> +    mSamples -= samplingFreq;

Is length ever > 2*samplingFreq? If so then you fall behind here.

Maybe a while(>) -=, an assert, or mSamples %= samplingFreq?

Comment 5

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

https://hg.mozilla.org/integration/mozilla-inbound/rev/cb894b5d342f

Comment 6

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

Backed out for perma-orange on b2g emulator M10 unrelated to this patch
https://tbpl.mozilla.org/?tree=Mozilla-Inbound&rev=cb894b5d342f

http://hg.mozilla.org/integration/mozilla-inbound/rev/c0eaa844f2f0

Comment 7

[Wes Kocher (:KWierso) (Not reading bugmail; email directly if needed)]

https://hg.mozilla.org/mozilla-central/rev/cb894b5d342f

Comment 8

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

Backed out of central:
https://hg.mozilla.org/mozilla-central/rev/ac6cbaa47f34

Comment 9

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

https://hg.mozilla.org/integration/mozilla-inbound/rev/daf5df0306b2

Comment 10

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

backed out:
https://hg.mozilla.org/integration/mozilla-inbound/rev/206169eef995

Comment 11

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

https://hg.mozilla.org/integration/mozilla-inbound/rev/78de1d480f29

Comment 12

[Ryan VanderMeulen [:RyanVM]]

https://hg.mozilla.org/mozilla-central/rev/78de1d480f29