A Music Generation Model Based on Generative Adversarial Networks with Bayesian Optimization

In recent years, a huge number of neural networks have been applied in music generation, many of which use generative adversarial networks (GAN). In this paper, a novel melody generation framework is proposed to create motivation for composers, which contains a generator made by bidirectional long short-term memory (Bi-LSTM) and a discriminator made by long short-term memory (LSTM). We change the traditional optimization policy of GAN by bringing Bayesian optimization in our model. In last, we conduct a user study that show better preference of our generated melodies over that produced by several recent other music generation models.