Paper short abstract:

Emerging research on vocal biomarkers for autism diagnosis mobilizes machine learning to enact the ‘autistic voice’ as an entity rooted in individual biology. How might histories of queer voices unsettle current attempts at making minoritarian identities legible through machine listening?

Paper long abstract:

Following technological developments in the field of artificial intelligence and ubiquitous computing, human voices are increasingly cast as a repository of rich information to be extracted through machine learning. In the medical field, the quest for ‘vocal biomarkers’ exemplifies this notion of the human voice as a correlate of, and an avenue into, a range of physio- and psychopathological states. Research on digital biomarkers, of which vocal biomarkers are a sub-type, aims at establishing a direct link between ‘the biological’ and ‘the digital’ by biologizing both the digital traces left behind by individuals, and the physio- and psychopathological conditions they are supposed to stand in for. Although no vocal biomarkers have been approved for clinical use yet, major investments are being made in their discovery, and they are being mobilized by medtech startups. This presentation asks what a voice can (be made to) do when enhanced through big data and machine learning. Specifically, it centers on research on vocal biomarkers for autism diagnosis. Targeting autism as a diagnosis troubled to this day by its lack of biomarkers, this emerging field of research mobilizes machine learning to enact the ‘autistic voice’ as an entity with stable and unchanging characteristics across individuals. Reading the history of the autistic voice, and its recent machine learning-fueled developments, in parallel to past and present attempts at biologizing queer voices, this presentation speculates on how queer theory and lavender linguistics might unsettle contemporary attempts at making minoritarian identities legible through machine listening.