Perception of speaker sex in re-synthesized children's voices

Abstract

Recent studies have shown that fundamental frequency (F0) and average formant frequencies (FF) provide important cues for the perception of speaker sex. Experiments on vocoded adult voices have indicated that upward scaling of F0 and FFs increases the probability that a voice will be perceived as female while downward scaling increases the probability that the voice will be perceived as male. The present study extends these manipulations to children’s voices. Separate groups of adult listeners heard vocoded /hVd/ syllables spoken by five boys and five girls from 14 age groups (5–18 years) in four synthesis conditions using the STRAIGHT vocoder. These conditions involved swapping F0 and/or FFs to the opposite-sex average within each age group. Compared to the synthesized, unswapped originals, both the F0-swapped condition and the FF-swapped condition resulted in lower sex recognition accuracy for the older females but relatively smaller effects for males. The combined F0 + FF swapped condition produced the largest drop in performance for both sexes, consistent with findings indicating that a change in F0 or FFs alone is generally insufficient to produce a compelling conversion of speaker sex in adults. [Hillenbrand and Clark, Attention Percep. Psychophys. 71(5), 1150–1166 (2009).]