Voice quality: Listener identification of African-American versus Caucasian speakers

Abstract

Although relative contributions of dialect and voice quality are difficult to tease apart, previous research has shown that listeners can perceive the gender and ethnicity of speakers at better than chance performance from recorded speech samples. The present study attempted to mitigate dialect cues and measure the resulting effects on perception by screening for dialect and by varying stimulus length. In experiment I, passages produced by ten African-American males, ten African-American females, ten Caucasian males, and ten Caucasian females were presented to 40 listeners with the same gender/ethnic distribution. In experiment II, sentence and citation /hVd/ stimuli produced by five of the speakers from each of the gender/ethnic groups were presented to 40 listeners. Both experiments required listeners to categorize the speakers by gender and ethnicity. Listeners were able to identify speaker gender and ethnicity at far better than chance performance. Accuracy of identification increased with stimulus length. Accurate identifications were made even in the citation condition, suggesting a unique contribution of voice quality apart from dialect. Caucasian male speakers were most accurately identified, while African-American females were least accurately identified. No listener gender/ethnic group performed more accurately than another. [Work supported by NIDCD.]