I've built an abstraction that glitches mp3 files.
It uses some of the concepts I discovered when I was working on the jpeglitcher. After reading some documentation about the mp3 file structure, I tried to find the "sweet spots" of these files: the places where a single byte change causes the greatest possible sonic mutations.
An mp3 file is made of a succession of frames. One frame is 26,122449 ms of audio, whatever the bitrate, frequency... After a period of trials and experimentations, I realized that the best places to glitch within these files where the so called "side information" area and the audio data itself. Messing around with frame headers, as I did in the beginning, was quite hazardous: files would easily break and, if readable, they just sounded too "destroyed" and sufferd heavy time domain damage.
I think about this abstraction as a synthesis tool: a unique way for producers to process drum loops (for instance) or any other part of a song. If you use the random number glitch, each instance of glitch of the same, unalterd original file will give you a different result. Moreover: a glitched mp3 will sound differently in every software player: vlc will have its own interpretation, probably very different from Itunes or any other DAW sequencer you import this file into...
The mp3glitcher is a CPU hungry abstraction when it comes to the actual glitching process. A typical 4'00" song can take up to 90 seconds to process on an old computer (but far less on more recent systems...) so don't panic if you don't get a result immediately. Also do not use this on a variable bitrate file as (for some reason beyond my comprehension) it will most likely not work.
Please feel free to use this tool and give me comments about it, so I can make it better.
Tested on mac (OS 10.5 and 10.9) and PC (win. xp/7). recquires pd-extended. The zip files is the stand alone version for mac