Comparing segmental and suprasegmental features in speaker gender perception: An acoustic distance approach

Abstract

Segmental (e.g., articulation) and suprasegmental (e.g., intonation) speech features provide meaningful perceptual cues to speaker gender. Yet, the relative perceptual importance of these features remains poorly understood. Further, proposed methods for quantifying these features often require labor-intensive measurement by human judges. This study used an acoustic-only method to quantify segmental and suprasegmental features among 60 speakers with varied gender identities (e.g., non-binary, transgender and cisgender men and women) and compared these acoustic measures to human perceptual judgements of speaker gender. Thirty listeners rated confidence in speaker gender and masculinity/femininity for each speaker. To obtain acoustic distance among speakers, three conditions were created that manipulated which acoustic features were present in the speech utterances. Acoustic distance was calculated in each of the three conditions between each speaker and a cisgender man and woman who were perceptually rated as most gender prototypical. Acoustic distances for the three conditions were fit to regression models to determine which was most predictive of listeners’ perceptual judgements. Both segmental and intonation features predicted perceptual measures. However, segmental cues better explained listener’s ratings, suggesting their greater perceptual weighting in listeners’ judgements of speaker gender.