i was thinking of doing it via a painful manual process of sampling a pattern, pitch shifting the sample by 8 or 10 octaves, then substitute the resulting sound for the original sounds in the pattern. possibly repeating as required. i'm sure there will be aliasing issues to sort out. maybe pitch shift wont be good enough - some sort of peak detection would do it?

figuring out an algorithm to do it in realtime is probably harder than hard, but a sample and hold type effect that dynamically adjusts the timbre of the sound based on the recently played rhythms could be crazy crazy sounding.