CCE Theses and Dissertations

Evolutionary Algorithm for Generation of Air Pressure and Lip Pressure Parameters for Automated Performance of Brass Instruments

Date of Award

2006

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Graduate School of Computer and Information Sciences

Advisor

Sumitra Mukherjee

Committee Member

Gregory E. Simco

Committee Member

Junping Sun

Abstract

The artificial mouth is a robotic device that simulates a human mouth. It consists of moveable lips and an adjustable air supply. The uses of an artificial mouth include research for physical modeling of the lips and automatic performance. Automatic performance of a musical instrument is when an instrument is played without the direct interaction of a human. Typically mechanics and robotics are used instead of a human.

In this study the use of a genetic algorithm to compute air pressure and lip pressure values so that the artificial mouth can correctly play five notes on a brass instrument is investigated. In order to properly playa brass instrument, a player must apply proper tension between the lips and apply proper airflow so that the lips vibrate at the proper frequency. A player changes the notes on a brass instrument by depressing keys and changing lip pressure and air flow. This study investigated a machine learning approach to finding lip pressure and air pressure parameters so that an artificial mouth could play five notes of a scale on a trumpet. A fast search algorithm was needed because it takes about 4 seconds to measure the frequency produced by each combination of pressure parameters. This measurement is slow because of the slow moving mechanics of the system and a delay produced while the notes are measured for pitch. Two different mouthpieces were used to investigate the ability to adapt to different mouthpieces. The algorithm started with a randomly generated population and evolved the lip pressure and air pressure parameters with an evolutionary algorithm using crossover and mutation designed for the knowledge scheme in this application. The efficiency of this algorithm was compared to an exhaustive search. Experimentation was performed using various combinations of genetic parameters including population size, crossover rate, and mutation rate. The evolutionary search was shown to be about 10 times faster than the exhaustive search because the evolutionary algorithm searches only very small portion of the search space. A recommendation for future research is to conduct further experimentation to determine more optimal crossover and mutation rates.

This document is currently not available here.

  Link to NovaCat

Share

COinS