Music from BALA: A Music Generation System

Abstract

We present a system which uses a Long Short Term Memory Generative Adversarial Network (LSTM-GAN) for building a model for Music generation and web application to interact with the model. We choose GANs since it is the efficient way of training deep generative neural networks. For the music generation from this model, we will be training the model on a collection of different music datasets and we will build the web application to listen to model generated music. The success of this project can help in different ways such as our system as a Sparring Composer i.e when Music composer goes out of ideas, our system can be used for practice to get insights for the generation of new music composition, using our system to generate functional music like musics for advertisement and relaxing music. Based on the trained model we develop the web app where user can listen AI powered music of different flavour. Our model can be trained on different genres of music for the exciting outputs.