A Technological Framework to Teach Music Online via Machine Learning with the Focus on Automated Chord Detection

Abstract

This thesis proposes possible approaches to adding technology to music courses and examines the impact online music courses may have on music teaching and practicing. The main question in this research is: How can we build an artificial system that, improves its ability to sense and coordinate with human musician as they are learning and practicing music? Artificial Intelligence and Human-Computer Interaction have enhanced computer music systems’ capability to perform with humans through a broad spectrum of applications. However, musical interaction between humans and current applications is still less musical than the interaction between actual humans. This thesis incorporates various techniques, especially machine learning and deep learning algorithms, to make the experience of learning and practicing music more intuitive and efficient for music lovers. The current system covers three fundamental aspects of human-computer collaborative music performance and practice: 1) music theory curriculum, 2) chord detection, and 3) score following. We developed a Web-based prototype to teach people with little to no music background the music theory basics and how to play the piano. To address this problem, we developed our prototype in two phases. The first one is for students to learn about the basics of music theory. Secondly, students will be able to practice what they learned by playing their favorite songs on the application which connects to their own digital or acoustic piano and provides proper guidance. We implemented a model that trains a different set of parameters based on each individual measure and focuses on predicting the number of chords and notes per chord. A nearest- neighbor search algorithm will decode an improvised score to select the training example closest to the estimation given the model prediction.