This item is accessible only to the Washington University community.

Off-Campus WUSTL Users: Click the “Off-Campus Download” button below. You will be prompted to log in using your WUSTL Key.

Date Submitted

Spring 4-30-2014

Research Mentor and Department

Dennis Mell

Restricted/Unrestricted

Dissertation/Thesis

Abstract

Many musicians improve by playing an active role in a band. Musicians also improve by learning how to transcribe and play songs on their own. If a musician could isolate an instrument in a song recording, he or she could either learn the instrument's part or omit that part from the recording in order to play with the remainder of the band. This project was aimed at isolating instruments to allow a musician to quickly improve through these learning techniques. The main goal was to take a prerecorded audio track – consisting of multiple instruments playing together – and separate it into several tracks of only one instrument each. The first attempt at this goal used non-negative matrix factorization, which factors a recording’s soundwaves (stored as a matrix of numbers) to it split up into single-instrument tracks. Another technique, which involved finding peaks in the amplitudes of dominant pitches in recordings and isolating them, proved to be more robust. While this technique had certain shortcomings, it was successful in separating different pitches and storing them on separate tracks. Perhaps by adding functionality to this method, it may be possible to create an algorithm that can identify the instruments in a recording and separate them into tracks accurately.