LARA-Gen: Enabling Continuous Emotion Control for Music Generation Models via Latent Affective Representation Alignment
Published in arXiv preprint, 2025
We propose the Latent Affective Representation Alignment mechanism for continuous fine-grained emotion control in music generation models. The method accepts continuous valence-arousal values as input, effectively decoupling emotional attributes from music content, and significantly outperforms baselines in both emotional accuracy and generation quality.
Recommended citation: Jiahao Mei, Xuenan Xu, Zeyu Xie, et al. "LARA-Gen: Enabling Continuous Emotion Control for Music Generation Models via Latent Affective Representation Alignment." arXiv, 2025.
Download Paper