Using virtual instruments, also mixed with public domain recordings (like URMP), no gen ai. Using the ~30 sound banks I have I use a method I published (check out the soundfont babies blog) for mixing them together. For cc recordings I split and stretch them to play target midi notes.
The full benchmarks can be checked out on the GitHub. The strongest model point is onset - which is similar to pitch. The weakest being 'offset', which is length. Both scores still edge out Mirelo's closed source model, and significantly beat MuScriptor.
I haven't tried anything outside of 12TET, but probably not good (it predicts in 12TET). Alternative tunings should work fine (it re-tunes to a440 for the model).
Not a legal expert, but that sounds like a trojan horse. Does this mean if a bad actor were to use Meta's model, that if they bypass the safegaurds that they get off scot-free?
So.. I feel like maybe it's not the best idea to let a known reward-hacking model to go rogue, especially when you can't truly see its chain of thought.
reply