TY - GEN
T1 - Symbolic Music Generation with Adaptive Representation Alignment in Texture
AU - Chen, Qing
AU - Zhang, Hengyu
AU - Liu, Kaiyuan
AU - Lu, Tianyi
AU - Zhou, Qidang
AU - Dong, Daoguo
AU - He, Liang
N1 - Publisher Copyright:
© 2025 ACM.
PY - 2025/10/26
Y1 - 2025/10/26
N2 - Symbolic music is gaining more and more attention for its high semantic precision and editorial flexibility. However, using traditional condition processing methods to generate symbolic music is problematic since these approaches often result in semantic loss by overlooking the intrinsic connections between musical elements. We introduce the SMART, Symbolic Music generation with Adaptive Representation alignment in Texture which incorporates a condition disentanglement method and an adaptive representation alignment framework, capable of fully utilizing multi-scale information and generating music with consistent texture. In SMART, the condition disentanglement method is used to decouple complex condition into more refined musical elements and the representation alignment framework is for semantic constraint during the generation process. Our method demonstrates remarkable advancements in texture consistency and chord controllability compared to standard strong baselines in symbolic music generation.
AB - Symbolic music is gaining more and more attention for its high semantic precision and editorial flexibility. However, using traditional condition processing methods to generate symbolic music is problematic since these approaches often result in semantic loss by overlooking the intrinsic connections between musical elements. We introduce the SMART, Symbolic Music generation with Adaptive Representation alignment in Texture which incorporates a condition disentanglement method and an adaptive representation alignment framework, capable of fully utilizing multi-scale information and generating music with consistent texture. In SMART, the condition disentanglement method is used to decouple complex condition into more refined musical elements and the representation alignment framework is for semantic constraint during the generation process. Our method demonstrates remarkable advancements in texture consistency and chord controllability compared to standard strong baselines in symbolic music generation.
KW - diffusion model
KW - representation alignment
KW - symbolic music generation
UR - https://www.scopus.com/pages/publications/105028934077
U2 - 10.1145/3746278.3759377
DO - 10.1145/3746278.3759377
M3 - 会议稿件
AN - SCOPUS:105028934077
T3 - McGE 2025 - Proceedings of the 3rd International Workshop on Multimedia Content Generation and Evaluation: New Methods and Practice, Co-Located with MM 2025
SP - 12
EP - 18
BT - McGE 2025 - Proceedings of the 3rd International Workshop on Multimedia Content Generation and Evaluation
PB - Association for Computing Machinery, Inc
T2 - 3rd International Workshop on Multimedia Content Generation and Evaluation: New Methods and Practice, McGE 2025
Y2 - 31 October 2025 through 31 October 2025
ER -