Short abstract:

With the rise of AI diagnostic methods, the voice became a site revealing the disturbances in the mind and body unavailable to human consciousness. Voice biomarker technology could be framed as an aesthetic agent that makes aesthetic and moral judgments revealing new body-mind-machine relations.

Long abstract:

Compared to other medical fields, psychiatry was falling behind in developing objective quantifiable measures (Barron, 2021). This has changed with the rise of computational and AI methods in psychiatry. Advocates of digital phenotyping in mental health proposed voice as a generative site for detecting symptoms and signs of distress (Fagherazzi et al., 2021). Complexity and valence of words used, pause length, intonation, rate of vocal fold vibration and other markers are used to determine mood fluctuations, distress, and the onset of a clinical condition.

Vocal biomarker analysis has been mostly critiqued as a successor of discriminatory scientific practices and universalising concepts (Ma et al., 2023; Semel, 2020). In this paper, instead of viewing it as a controversial tool, I argue that vocal biomarker-detecting algorithms are aesthetic (relating to sensible experience) and moral agents. In labelling and training, algorithms acquire aesthetic skills to perform aesthetic judgments. Sensors render perceivable, organise, categorise, and relate sensory knowledge about how the voice “appears” to non-human listening. Algorithms thus produce “hyper-aestheticised articulations” (Fuller & Weizman, 2021) of mental health. In addition, as mental health always carries a moral dimension (Vidal & Ortega, 2019), algorithmic judgements of (predicted) deviance, responsibility, and corrective actions assign normative values to the expression of affect, behaviour, and (ill) health.

The aesthetic lens provides a generative view on the emerging body-mind-technology relationships in digital mental health and transformations in the expression and experience of mental health. It also provides opportunities for alternative design models for AI for vocal biomarkers detection.