“A sociolinguistic perspective on why it is necessary to distinguish among types of creaky voice”
Robert J. Podesva
As discourses about creaky voice started circulating in the media a decade ago, scholars have shed considerable light on its sociolinguistic complexity (Podesva 2013, Lee 2015, Becker et al. 2017, Zimman 2017). This paper, by providing evidence from two studies, argues that attending to the phonetic complexity of creaky voice (Keating, Garellek, and Kreiman 2015) offers a clearer picture of its social patterning and indexicality.
The first study examines voice quality variation in a large corpus of spontaneous California English to evaluate the empirical validity of social discourses that characterize young women as the quintessential producers of creaky voice. Crucially, creaky voice is quantified in a variety of ways – through binary classification as well as acoustic measures that each capture distinct phonetic properties. Results indicate that even though women exhibit a higher incidence of creak than men phrase-finally, the gender difference disappears in earlier positions in the phrase, a prosodic domain into which younger speakers have introduced creak. Further, while women exhibit phonetically stronger creak than men overall, young men and the oldest women rival younger women in this regard. That young men’s patterns resemble those of young women suggests that creaky voice presents indexical potentials that go beyond gender.
The second study investigates voice quality variation in an audiovisual corpus of dyadic interactions among young people in Northern California to probe the social indexicality of creaky voice. Previous discourse analytic studies (Grivičić and Nilep 2004, Lee 2015, Zimman 2017) have independently claimed that creaky voice indexes negative affect or disengagement. This study tests whether these claims can be generalized and also attempts to disentangle evaluative stance (negative affect) from footing/alignment (disengagement). Speech classified as creaky was grouped into two clusters on the basis of 4 acoustic measures that capture different phonetic dimensions of voice quality. One cluster exhibiting the acoustic properties of what Keating et al. (2015) call “prototypical creak” was more likely to include words conveying lower arousal (based on computational approaches to sentiment analysis) and was characterized by less body movement (based on computer vision analysis). A second cluster with the acoustic properties of what might be more accurately labeled “vocal fry” was more likely to include words with negative valence and phrases with more negative sentiment. In conclusion, acoustically distinct qualities of creaky voice (prototypical creak and vocal fry) conventionally index distinct interactional properties (disengagement and negative affect, respectively). Teasing them apart would not be possible without making finer grained distinctions among vocal practices that all fall under the domain of “creak.”
ONLINE (Zoom link: https://qmul-ac-uk.zoom.us/j/87971879921?pwd=Z1o3V3crVWlIekR1MXBUNmI2Mm9xQT09)