What I'm working on is a markov model for sound creation. Unlike previous models which were interested in notes and midi genereation, this works by looking at the actual samples in the audio file.
Right now: mixed results with low order & short sound files. fills up memory with higher order & larger sound files.
Obviously this is huge in terms of resource consumption and the project is pretty much on hold pending some time to sit down and do a c implementation