Effects of gender neutralization on the anthropomorphism of natural and synthetic voices

Abstract

This study examines the impact of gender neutralization on the anthropomorphism of speech signals and is motivated by the need to find a computer-generated voice for our android robot Andrea. A filter was used to gender-neutralize recordings from natural and synthetic voices, which were then pre-tested for gender neutrality. The results of the main experiment showed that gender-neutral voices were less anthropomorphic than gender-specific voices, with the naturalness of the voice having a greater impact than gender neutralization itself. Synthetic voices were rated less anthropomorphic than natural voices, and the additional effect of gender neutralization was stronger for natural voices. Overall, anthropomorphism ranked highest for natural gender-specific voices, followed by natural gender-neutralized voices, synthetic gender-specific, and, finally, synthetic gender-neutralized voices. Gender of the participants had no significant impact. These findings have implications for the development of humanoid and social as well as android robots.