When generating symbolic music, people sometimes encounter problems such as unclear structure or mismatch between tracks.Accordingly, this study proposes a multi-track symbolic music generation method to solve the problems of melodic fragmentation, uneven rhythm, and multi-track conflicts in traditional music generation.Experiments on several music datasets show that compared with other benchmark music generation models, the generated music has a pitch class histogram similarity of 0.91 and a pitch distance of 1.28.The expert evaluation report shows that the structural consistency score is 4.52, and the melody fluency score is 4.56.The synchronisation rate of multi-track rhythm reaches 92%.These findings show that the introduction of structural constraints and cross-track cooperation is effective and can generate more logical and coherent music.This study verifies the feasibility of diffusion model in the generation of symbolic music, and provides a practical multi-track composition method for researchers in other music-related fields.
Wang et al. (Thu,) studied this question.