Music Generation Using Deep Learning

Abstract

In this paper, we explore the usage of char-RNN which is special type of recurrent neural network (RNN) in generating music pieces and propose an approach to do so. First, we train a model using existing music data. The generating model mimics the music patterns in such a way that we humans enjoy. The generated model does not replicate the training data but understands and creates patterns to generate new music. We generate honest quality music which should be good and melodious to hear. By tuning, the generated music can be beneficial for composers, film makers, artists in their tasks, and it can also be sold by companies or individuals. In our paper, we focus more on char ABC-notation because it is reliable to represent music using just sequence of characters. We use bidirectional long short-term memory (LSTM) which takes input as music sequences and observer that the proposed model has more accuracy compared with other models.