AI for changing genre of already existin music

Most of current available AI models use audio music and separate voices into different tracks and that is ok. Next step would be that each separated voice is converted into notes (like score or sheet music) with proper metric, tempo and so on.

After that step there is missing AI which would completely do analise of these generated notes (audio is converted into notes, similarly as if audio speach would be converted into words and all of next processing is of those words and not audio) like profesional composers do.

So such AI should be trained by tons of sheet music scores from Bethoween to modern classics, but also with arangements of different rythms, phrases, manuvers, different dynamics performed on different instruments like picikato, stacato, forte, legato and so on.

So the goal would be that generated notes (as XML, or MIDI or just scores in pdf) are transformed into other notes (also in pdf or xml or midi) but this time the represent the same song but in different genre, lets say classic, ir POP, or rock or whatever.

In that transformation, vocal track from original song, should be cut into parts and rearanged to properly fit into possible new metric of changed genre, and new audio track should be "rendered" from those new generated notes, including full symphonic orchestra if that is target genre, or added new styles or instruments which were not existed in original song but exist in target genre.

Harmony of the original song should be preserved, however it can be advanced into rules characteristic for the new genre (for example jazz). So the ultimate goal of such AI is notal transformation from notes extracted from original song to notes of targeted genre for all notal tracks and final audio render of those notes with realistic sound of instruments and final remix with some of originally extracted instrument/voices