Eunjin Choi, Hounsu Kim, Hayeon Bang, Taegyun Kwon, Juhan Nam
D3PIA is a discrete diffusion model that effectively generates piano accompaniments from lead sheets, outperforming other models in preserving chord conditions and musical coherence.
Creating piano accompaniments from lead sheets, which include melodies and chords, is a complex task in music generation. The researchers developed D3PIA, a new model that uses a method called discrete diffusion to generate these accompaniments. By focusing on the local relationship between the lead sheet and the accompaniment in a piano-roll format, D3PIA improves the accuracy of chord representation and the overall musicality of the generated pieces. Tests showed that D3PIA outperforms existing models in maintaining chord integrity and producing music that sounds more coherent to listeners.
