A Review of Research on Emotional Music Generation

Abstract

With the expansion of the development field of artificial intelligence, emotional music generation’s potential in the field of music generation has been gradually explored. In this paper, rule-based, deep learning and reinforcement learning methods for emotional music generation are analyzed, respectively. Rule-based method generates music according to certain theories and patterns, which is interpretable but has a single style. Deep learning generates music by learning the characteristics of music data through neural network models, which have rich emotional levels but rely on a large amount of data and have poor flexibility. Reinforcement learning allows agents to optimize behavior strategies in interaction with the environment to generate music, which is flexible but complex. Through comparative analysis, it is found that the rule-based method can be used as a basic framework to constrain other methods, deep learning has outstanding performance in creating diversity and generation quality, and reinforcement learning has obvious advantages in generating specific emotions. The integration of the three methods is expected to further improve the effect of emotional music generation and provide strong support for music creation.