"Separating Sounds Using Genetic Programming in Matlab"

Supervised by G.Brown

Abstract

For obvious reasons it is desirable that a user is able to speak to their computer and the computer is able to understand and interpret the speech. There are currently several speech recognition software packages commercially available that claim to allow just that. While they are all reasonably successful in a quiet environment they all develop serious problems if there is any background noise. If the noise could be removed before speech recogniser attempted recognition it's performance would be dramatically improved. There are other uses for a method sound separation as well. If a recording is made of a musical performance but someone coughs close to the microphone during the performance then removing the sound of coughing without damaging the sound of the music would be desirable. Sound separation could also be used on modem lines, capacity would be increased if crackle were removed without changing the signal being transmitted.