Acoustic Differences Between Gay And Straight Male Speakers Of Australian English

Objectives: Variation in speech sounds conveys socio-indexical information regarding speaker characteristics and group affiliations, including sexual orientation. In perception, listeners can reliably identify speakers’ sexual orientation as gay or straight from speech samples in several languages, including Australian English. Listeners are reported to rely on higher mean F0 and range, and higher M1 in /s/ to identify gay men; however, studies on acoustic differences between gay and straight men show inconsistent results. Some studies report that gay men produce speech with higher mean and range in F0, as well as higher M1 in /s/ relative to straight men. Other studies found no correlation between sexual orientation and acoustic features, and only a few studies explored gay-straight differences in Australian English.

Methods: We collected data from 471 gay, bisexual, and straight male participants as part of a larger on-line survey on gender expression, sexual orientation, and mental health of Australian men; as part of the survey, participants submitted a short speech sample recorded via their smart phone devices. Speech samples were pre-screened for audio quality and accent-nativeness to select Australian English-speaking participants. Sixty-two participants (gay = 34; straight = 28) were selected for acoustic analysis. Eight vowels per participant were identified for F0 analysis; F0 was extracted as mean F0 in the mid-50 ms of each vowel. For each participant, F0 range was calculated as the standard deviation (SD) of F0 across the eight vowels. Two /s/ tokens per participant were identified. M1 was extracted from 50%-75% of the total /s/ duration. Mean F0 was modelled using generalised linear mixed models with the fixed effect Sexual Orientation, random effects for Speaker and Token, and log-normal family, as it showed right-skewed distribution. SD F0 was modelled using linear mixed models with the fixed effect Sexual Orientation and log-normal family; no random effects were used as the data were aggregated. M1 was modelled using generalised linear mixed models with the fixed effect Sexual Orientation, random effects for Speaker and Token, and gamma family, as it showed right-skewed distribution.

Results: M1 of /s/ is significantly higher for gay compared to straight men (β = 781.98, std. error = 82.44, t-value = 9.485, p < 0.001). Our results showed no significant differences between gay and straight men in mean and SD of F0.

Conclusion: Our results seemed to indicate that gay speakers use phoneme-level markers to express their identities, rather than voice characteristics.