Falsetto Detection Using Gaussian Mixture Models

Music 220C, Spring 2006

Gautham J. Mysore

This project will form one part of a larger project (which is being done with Ryan Cassidy). The main project is to develop an intelligent audio effect which processes a singer's voice only when the singer goes into falsetto. The first part of the project is to develop a falsetto detection algorithm which will determine if the singer's voice is in falsetto or not. This is the part that I will be work on this quarter. The concentration will be on machine learning and I will use Gaussian Mixture Models.

The other part of the larger project is the actual processing of the singer's voice if the singer is in fact in falsetto (as determined by the falsetto detection algorithm). If the singer is in falsetto, it will process the voice to simulate a pitched scream by adding a sense of roughness to the voice. This would be useful particularly for certain singers in the rock music genre, as it would allow them to go well above their standard vocal range while still maintaining the rough timber of their voice.