Hey, just popping in for a bit.
Bren, interesting thread. However, I'm not sure of the methodology in your experiment- hear me out.
***Warning: somewhat technical***
MP3, like all modern compressed audio formats, relies on the imperfections of the human ear, significantly in the phenomenon of 'masking'. That is, a tone of a certain frequency may 'mask' weaker tones around it. Think of an MP3 encoder calculating a spectrum graph every 1/75 of a second, finding the spectrum peaks, and marking all nearby weaker tones as 'masked' (the ear ignores these in the presence of the strong tone). The MP3 codec throws these masked tones out.
Fast forward to doing a waveform difference on compressed vs uncompressed audio: part of the difference will be these 'completely masked' tones, but as the dominant, masking tone is gone (since it's the same), you'll hear these otherwise-invisible sounds.
Likewise, if the psychoacoustic model MP3 uses is wrong for your ears, just a little difference in the waveform may unmask borderline tones or mask otherwise-audible tones.
So, neither of the two 'obvious' conditions hold: either that a small difference in waveforms strongly implies that the two waveforms will sound the same to a given person, or that a big difference in waveforms strongly implies that the two waveforms will sound different to a given person.
Counter-intuitive, I know. Blame our ears.
RD
