MeloDance: Dance Generation Guided by Music Structure and Emotion
Yixuan Li, Qiang Jin, Huaping Liu, Jinhai Chen, Xiangyu Zhao, Peng Li · 2025
This paper focuses on music-driven dance motion generation, in which current methods often face challenges such as irregular dance movements and the inability to capture the relationship between music and dance emotions. To address these issues, we propose MeloDance, a new dance generation framework, which includes two main modules. The Musical Structure Feature Analysis (MSFA) module is introduced to enhance the consistency of dance movements by considering the global structural features of music. The Emotion Alignment Module for Audio-Dance (EAM-AD) is designed to ensure that the emotional expressions of the movements match those of the music. Through experimental comparisons, our approach outperforms previous methods, generating more harmonious dance movements. Specific examples are available at: https://pkjq11.github.io/melodance-demo/