Simple Score Follower: A Contextual Switching Approach to Polyphonic Score Following on the Web using Deep-Learning Pitch Detection

Abstract

In this thesis we propose a contextual switching approach that combines modern deep-learning-based pitch detection technology with chroma-based chord recognition methods for the purpose of real-time, polyphonic score following based on live audio input from a mobile device. The goal of this experiment is to test the viability of a score following algorithm that switches between a pitch detection subroutine and a chord recognition subroutine based on the expected input at any time during a performance of an arbitrary musical score. The proposed solution is integrated into a web-based, cross-platform application prototype that can run on any laptop or mobile device to serve as an accessible tool for musicians. We begin with an overview of common methods used in score following, pitch detection and chord recognition. We then describe the design, development, and iterations of our proposed solution. We conclude with an assessment of the comparative performance of our application prototype against a commercially available mobile application with effective polyphonic score following capabilities.