I've discovered a problem with the AudioAvailable code in nsBuiltinDecoder::AudioLoop(): I think this code [1]: audioDuration += PlaySilence(static_cast<PRUint32>(missingSamples), channels, sampleTime); Should be: audioDuration += PlaySilence(static_cast<PRUint32>(missingSamples), channels, playedSamples); At that call site, we're injecting missingSamples samples of silence between the end of the last sound sample, and the start of the next chunk of sound samples. sampleTime is the number of samples which the *next* queued SoundData's audio starts at, but we need to inject missingSamples samples in order for playback to reach that position. Thus sampleTime is actually the *end* of the silence chunk which PlaySilence() will inject here, not the *start* as expected by PlaySilence(). playedSamples is the start of the silence chunk (i.e. the end of the last played chunk). [1] http://mxr.mozilla.org/mozilla-central/source/content/media/nsBuiltinDecoderStateMachine.cpp#440

Fixed in bug 598812.