Here's something I've been wondering for some time, but been too lazy to find out: Most lossless audio formats encode the data for each frame by first making a rouch approximation of the waveform, and then compressing the difference between the approximation and the real data. What occurs to me is, why not use MP3 for the approximation instead of a much simpler model? I suppose it would be easy enough to find out if there's any point in doing this...