The audio is 44.1khz stereo, but all of us use autoencoders so the songs will fit in a transformer's context window, and huge compression will affect quality. We're definitely working on better ones, though!