A Novel Method of Music Generation Based on Three Different Recurrent Neural Networks

manuscripts Nowadays, music generation has become an important research field in the study of deep learning. Many different methods such as Generative Adversarial Networks (GAN) and Variational self-encoding (VAE) are already proved to be able to handle such task. In this paper, RNNs, such as LSTM and GRU, will be used to compose the core of the network to predict notes of the melody and generate new music. MIDI will be used as the general music format because of its simple and appropriate data structure that can be coded as a dictionary.