Music Generation Using LSTM Neural Network

Abstract

With the advancement of machine learning, Neural Networks are utilized in various fields like music, writing and pictures. Music generation is a troublesome undertaking and has been effectively investigated since decades. In this paper we are proposing a system through which we can generate music automatically using Recurrent Neural Networks. The existing system like the markov model or graph-based minimization methods lack thematic structures and they are usually repetitive sequences of the same nodes. They typically overlook the data in the negative time heading, which is non-minor in the music expectation task, so we propose a bidirectional LSTM model to create the note succession. The system which we are proposing would produce unique and coherent compositions. We used a Long Short-Term Memory (LSTM) network. They are a kind of Recurrent Neural Network that can proficiently learn through inclination plunge. Utilizing a gating instrument, LSTMs can perceive and encode long haul designs. LSTMs are very helpful to take care of issues where the network needs to recall data for an extensive stretch of time similar to the case in music and text age.