A 2022 study of audio deepfakes at the University of Florida reverse-engineered the shapes of the vocal tracts of the AI-generated “speakers” using fluid dynamics.
The scientists discovered the vocal tracts of the deepfake voices had no correlation to human vocal tracts, even when a human listener was unable to distinguish the deepfake from the real voice.
Source: Blue, L. and Traynor, P. (2022) Deepfake audio has a tell – researchers use fluid dynamics to spot artificial imposter voices, The Conversation. Available at: https://doi.org/10.64628/AAI.9x57f5jvy.
John Kannenberg is an artistic researcher whose work investigates sounds as museological objects. Via an acoustemological approach, he considers the histories and cultures surrounding sounds, the technologies that generate or record them, and the auditors who hear or listen to them. He holds a PhD in sound & museums from the University of the Arts London, as well as a Master of Fine Arts in Studio Art degree and Graduate Certification in Museum Studies from the University of Michigan. He is Director and Chief Curator of The Museum of Portable Sound. Learn more about his work at johnkannenberg.com.
View more posts
