This article focuses on the application of DL in the field of music emotion recognition and intelligent generation. The purpose of this study is to build an efficient musical emotion recognition and intelligent generation model with the help of DL (Deep Learning) technology, and to solve the shortcomings of traditional methods in musical emotion processing. The research adopts a comprehensive music data set constructed by ourselves, covering over 5,000 music works with multi-style and multi-emotional annotations. By building a model combining CNN (Convolutional Neural Networks), RNN (Recurrent Neural Networks), GAN (Generative Adversarial Networks) and VAE (Variable Auto Encoders), and training in a specific experimental environment. The results show that the recognition accuracy of musical emotion recognition model is stable at about 90%, sad emotion is about 87%, passionate emotion is about 81%, and calm emotion is about 75% in the later training period. In terms of emotional consistency, the music generated by the music intelligent generation model has an average score of about 7.5 points in the later stage, 7 points in the sad emotion, 6.8 points in the passionate emotion and 6.5 points in the calm emotion. In quality assessment, the similarity with real music reaches about 0.67 after 150 rounds of training. Thus, the model based on DL shows good performance in music emotion recognition and intelligent generation.
Yanlin Liu (Sun,) studied this question.