We pay a lot of attention to audio quality. Our signal path takes the decoded audio as 16-bit integer or 32-bit integer or float samples, and converts to 32-bit float, then holds that as discrete channels.

When outputting, the channels are resampled to the rate of the output (if needed) and pitch shifted for playback speed, then the channels are mixed to their correct levels (as set by the clip definition) and mixed together.

As with video, each output is treated distinctly from data as unsullied as possible from the file, so no damaging gain structure changes are needed inside the audio chain, and each output gets audio as untouched as possible.