I'm asking myself if a synth is just a boring application for a model like this. But maybe it would be possible to implement a new kind of efficient sound compression algorithm.