MemoMusic 4.0: Personalized Emotion Music Generation Conditioned by Valence and Arousal as Virtual Tokens

MemoMusic aims at recommending or generating emotion music to meet the emotional requirements of individual listeners and navigate them to more positive emotional states. For MemoMusic 4.0, valence and arousal are applied as virtual tokens for emotion conditioned music generation. And predicting various music features is approached as multi-task learning. Additionally, adaptive weights are assigned to the loss of relevant features based on emotional conditions. Experimental results have validated the emotional regulation ability of MemoMusic 4.0 and the effectiveness of valence and arousal as virtual tokens for emotion music generation.