Novel LSTM-GAN Based Music Generation
Guangwei Li, Shuxue Ding, Yujie Li · 2021 13th International Conference on Wireless Communications and Signal Processing (WCSP) · 2021
With the rapid development of deep learning, many models for music generation have emerged. There are, however, many problems for methods based on the general neural network model, such as slow calculation speed, complex calculation, and long-term dependence. This study proposes a combined model method for music generation, in which the long short-term memory (LSTM) neural network and generative adversarial network (GAN) are combined to form an LSTM-GAN model. In this paper, a new data preprocessing conversion rule is proposed to process the musical instrument digital interface (MIDI) message data obtained by the performance coding method. Finally, the effectiveness of the proposed model by the maximum mean discrepancy assessment is verified. Experimental results demonstrate that the proposed model can produce novel music automatically and have good performance.