Music tutor : an evaluation tool for music performance

Abstract

In this dissertation a prototype music learning tool is proposed in order to aid beginner piano students in practising a piece of music in the absence of their teacher. The tool evaluates the students’ performance by comparing it with the teacher’s performance, using Music Information Retrieval algorithms which focus on the extraction of musical data from musical signals. These algorithms make use of multi-variate analysis such as Non-Negative Matrix Factorization (NMF) to detect the pitch of the music signal and time series analysis algorithms such as Dynamic Time Warping (DTW) to evaluate the student’s performance. In this respect, the musical signal is treated as a complex signal which can be analysed using standard signal processing techniques such as the Fourier Transform. The tool uses an interface which is designed to simulate the studentteacher environment by pointing out mistakes in the student’s performance and helping the student to correct his/her own mistakes. The algorithms were tested individually, where a high classification rate was obtained for both the detection and the evaluation algorithms. The prototype learning tool was also tested with beginner students, where the tool was observed to help the students in improving their performance of the music being studied.