Wow, that's comprehensive.
Count me excited as well. I'm very curious to see what can be done by assembling a bunch of different models. Maybe one model will convert a track exceptionally well, except for one part of the song ... but a different model will do that one part much better, but doesn't do the rest of the song as well. Then the results of the two conversions (or more) can be spliced together to create the best possible results.