Currently, we rely on a branch version of CSCore for all audio output on both Windows and Linux. The better solution would be to allow for more accurate sound output, especially once we start moving away from just red book audio.
See PortAudio as an example of a good cross-platform library.