Music Generation with Bi-Directional Long Short Term Memory Neural Networks

Deep Learning has been utilized in music generation where long term dependencies and patterns need to be learned by the network. Recurrent Neural Network (RNN) units have been used in the form of Generative Adversarial Networks, Auto-Encoders and Long Short-Term Memory Networks but they have been plagued by issues with regard to varying offsets and different sequence lengths. This paper proposes the modification of bi-directional Long Short Term Memory Networks to produce new music as well as generate the succeeding notes in existing compositions.