Send message to Author

Paper short abstract

This paper traces the genealogy of psychopathological voice analysis from Eberhard Zwirner’s Frequenzschreiber (1928) and Paul Moses’ Voice of Neurosis (1954) to contemporary AI vocal biomarker systems, asking which historical constructions of deviant vocality are encoded in today’s algorithms.

Paper long abstract

Contemporary voice AI systems that claim to detect depression, bipolarity, or neurosis from speech acoustics (Semel 2022; Low et al. 2020; Fagherazzi et al. 2021; Turow 2021) are routinely critiqued for encoding cultural, gendered, and ableist norms as universal biomarkers (Ma, Patitsas and Sterne 2023). What such critiques rarely pursue, however, is the historical depth of the epistemological formations they contest — the genealogy of techniques, measurement apparatuses, and normative assumptions through which voice first became a readable biometric index of psychic deviance.

This paper traces two foundational moments in that genealogy. First, Eberhard Zwirner's phonometric apparatus at the Kaiser-Wilhelm-Institut für Hirnforschung (Berlin-Buch, 1928–c. 1940), where the Frequenzschreiber translated the fleeting voice into quantifiable curves, operationalizing claims that speech rhythm and pitch encode psychiatric states — and, in the National Socialist context, racial-hygienic typologies. Second, Paul J. Moses' The Voice of Neurosis (1954), in which acousmatic analysis of recorded samples enabled diagnosis without clinical encounter, constituting voice as an autonomous biometric index of neurotic constitution and "androgynous" deviance.

Together, these cases illuminate how specific measurement technologies, institutional contexts, and normative assumptions about deviant embodiment were stabilized into diagnostic categories that persist — rearticulated in digital feature sets such as GeMAPS (Eyben et al. 2016) — in current clinical and commercial voice AI. Situating these histories within the panel's concern for historical legacies shaping biometric assemblages, the paper asks: which constructions of pathological vocality are encoded in today's algorithmic infrastructure, and with what political effects?