Maybe besides https://openai.com/blog/musenet/? The present solution is rather primitive; just a bit of random with time and scale quantisation; but some people seem to be quickly satisfied.