Learning to create piano performances

Nearly all previous work on music generation has effectively focused on creating musical scores. In contrast, we learn to create piano performances: besides predicting the notes to be played, we also predict expressive variations in the timing and musical dynamics (loudness). We have provided samples generated by our system to a set of professional musicians and composers, and the feedback was positive. Overall, the comments indicate that our system is generating music that, while lacking high-level structure, does indeed sound very much like human performance, and is closely reminiscent of the classical piano repertoire.