What I want to do is simply
mp3 = read_mp3(mp3_filename)
audio_left = mp3.audio_channels[0]
where audio_left will contain raw PCM audio dat
I am considering using FFmpeg as a subprocess. There is a Python wrapper called pyffmpeg, but I had difficulty installing it on my system (OS X 10.7.3).
You may also want to look at the code here for calling FFmpeg as a subprocess from Python: https://github.com/albertz/learn-midi/blob/master/decode.py