If it can do music, can we train better models for different kinds of music? Or different models for different instruments makes more sense? For different instr

If it can do music, can we train better models for different kinds of music? Or different models for different instruments makes more sense? For different instruments we can get better resolution by making the spectrogram represent different frequency ranges. This is terribly exciting, what a time to be alive.