Previous MSc Theses

"Using Recurrent Neural Networks to Generate Music." L. Devlin. D. Roussinov. Department of Computer and Information Sciences, University of Strathclyde. 2019. Download PDF (BibTeX) ILS

Abstract:
This project focuses on the application of Recurrent Neural Networks (RNN) to the field   of generative symbolic music. The following paper discusses the current state of RNN   music generation and proposes two representations for compositions with a shared   underlying data structure. Once the compositions had been generated, they were   compared against human-written compositions by using an empirical MIR analysis   system, as well as a subjective test in which participants offered perceptual responses   to the compositions. Overall, while the system was able to produce some melodically   pleasant results, it failed to capture the long term structure and complexity of human   compositions.