We present an array of generative machine learning models for the purpose of modeling and generating music. We also explore a number of different representations of musical data, some of which work significantly better than the others. Furthermore, we present a novel extension of traditional Markov chains for this specific task and analyze its performance quantitatively.
The project was undertaken as a part of the Fall-2018 Introduction to Machine Learning (10-701) course offered by the School of Computer Science, Carnegie Mellon University. The paper can be found here.