Youtube hosts lots of music transcoded using faac and quality isn't great. As popular tracks are usually uploaded by multiple users, one might reduce encoding noise by averaging waveforms of duplicates (can be done either in client app or intermediate server). Noise std should decrease as 1/sqrt(N:). Do you guys think this would work?
I hope this question is of a purely academic nature, because of course the right answer is if you want the music then buy your own copy.