User Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10_9_1) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/32.0.1700.77 Safari/537.36 Steps to reproduce: - call getUserMedia({audio:true,video:false}, Success, Failure); Actual results: After about 5 seconds, audio data is all 0 and the little green camera icon in the url bar disappears (indicating that firefox is not recording any more). Expected results: Shouldn't stop recording after 5 seconds.

You can see the bug here: http://webaudiodemos.appspot.com/pitchdetect/index.html Click on 'live input', then immediately click 'share selected device'. For some reason the bug only occurs if you click 'share selected device' quickly. If you wait a few seconds it seems to work fine.

Here is a workaround for anyone interested: https://support.mozilla.org/en-US/questions/984179

Paul -- This looks to be a Web Audio issue and not a WebRTC issue. Can I ask you to put your spec editor hat on and say if we're doing the right thing here or not? We're clearly not holding a strong reference here, and I think the question is: should we be? It looks like we should be, except I'm not sure if "connected in a cycle" is applicable here or not: https://dvcs.w3.org/hg/audio/raw-file/tip/webaudio/specification.html#lifetime-AudioNode If we aren't doing the right thing per the spec (and/or there's a bug in our implementation), I can find someone else to fix it (unless you prefer to). Thanks.

Same issue as bug 934512. Garbage collection should not be observable, so the objects producing sound should not be collected.

(In reply to Maire Reavy [:mreavy] from comment #3) > Paul -- This looks to be a Web Audio issue and not a WebRTC issue. Can I > ask you to put your spec editor hat on and say if we're doing the right > thing here or not? We're clearly not holding a strong reference here, and I > think the question is: should we be? > > It looks like we should be, except I'm not sure if "connected in a cycle" is > applicable here or not: > https://dvcs.w3.org/hg/audio/raw-file/tip/webaudio/specification. > html#lifetime-AudioNode > > If we aren't doing the right thing per the spec (and/or there's a bug in our > implementation), I can find someone else to fix it (unless you prefer to). > Thanks. Yes, as Karl says, this is a bug, and is already tracked.