Long Short-Term Memory Model Appliance for Generating Music Compositions

This paper represents an approach to the music generation based on a recurrent neural network. The key goal is to create a model, that can learn different musical styles, and then generate new compositions based on previously explored content. The model declares a function with the ability to remember the temporal state and short pieces of data to use them for future predictions. The deep-learned neural network uses a one-to-many bi-axial LSTM model trained with the convolutional kernel to generate a polymorphic music piece. A deep reinforcement learning (DRL) approach was used to encourage research and increase the global consistency of music compositions being created. For quantitative and qualitative analysis, this approach works well when generating polyphonic music.