This article describes a new method for generating music symbolically, using a graph that represents the structure of a song and a Graph Neural Network. The graph uses information such as note sequences and instruments as node features, while the correlation between note sequences forms the edge feature. This method allows for the understanding of the overall structure of multi-track music and facilitates the generation of each musical pattern for music generation. It also has potential applications in various fields in Music Information Retrieval, including music composition, music classification, and music inpainting systems.
Publication date: 29 Dec 2023
Project Page: N/A
Paper: https://arxiv.org/pdf/2312.15400
