If we underrun on the initial read (to fill the cubeb buffers) of AudioStream, it outputs X samples of real audio, then aFrames-X samples of 0's. This is not good and will cause a startup glitch. After uint8_t* rpos = static_cast<uint8_t*>(aBuffer) + FramesToBytes(aFrames - underrunFrames); we want something like this (mHasRun is a bool to tell us if this is the first time we've fed data from the DataCallback) Note: memmove, not memcpy! if (!mHasRun) { // underrun on initial read - output 0's *first*, then real audio to avoid glitch memmove(rpos, static_cast<uint8_t*>(aBuffer), FramesToBytes(aFrames - underrunFrames)); rpos = static_cast<uint8_t*>(aBuffer); }

This is somewhat handled by my work in progress patch for bug 919215.

Yes, this is handled in bug 919215, which has been fixed and is now on central.