realistic music generation is a challenging task. When building generative
models of music that are learnt from data, typically high-level representations
such as scores or MIDI are used that abstract away the idiosyncrasies of a
particular performance. But these nuances are very impor