Advancing generative AI for music with photonics

This paper presents PhotoSing, a system that learns to generate polyphonic tunes by extracting sequencing rules from given examples. We developed a method to encode given pieces of music in terms of unique musical events, referred to as UMEs, and stochastic rules for sequencing them. Those rules are subsequently converted into representations to be processed by a photonic computer to generate new compositions. This research builds upon a previous system, QuSing, which generated monophonic tunes with superconducting quantum computing. The paper discusses the pitfalls of the previous system, the research and the solutions developed to improve them. It details the system with demonstrative musical examples and analyses.