The University of Sheffield
Department of Computer Science

Sarah Wilson MSc Dissertation 2014/15

Characteristics of Robot, Alien and Cartoon Voices

Supervised by R.K.Moore

Abstract

In TV and film there are many alien, robot and cartoon characters with voice effects that make them sound out of this world. This project aims to research and present an overview of the effects and techniques that are used to alter voices. It aims to highlight which vocal characteristics are the most useful for measuring these effects and finally, it aims to provide a set of guidelines for how to alter a voice to suit the 'persona' of the character.

The voice data used in this project was source from TV programs, films and games. Voices are labelled as robot or alien and with their size, gender, material and personality. They are labelled with vocal measurements such as pitch, number of glottal pulses and voicing measurements and finally they are labelled with sound effects such as echo/reverb, harmony, modulation, speed changes or voice qualities. 

Results from principal component analysis found that measurements of noise in the voiced parts of the speech are what caused the most variance and that these measures correlated with pitch. Using linear discriminant analysis the voices of the character could be used to correctly classify the person with an accuracy ranging from 61% to 81%.