How to Find a Movement That Is Synced With Music

The current approach that we've used to find this movement that is synced is actually a fairly straightforward, just greedy search like approach. I would be very interested in doing this in a more machine learning fashion where we've learned mappings of given an input music. What are the characteristics that the output dance needs to have so that at test time, we can now just give an input piece of music, just directly predict what the movement should look like rather than having to run the search process at test time.