Brain processes deepfake voices differently than natural voices
People often accept deepfakes as natural. However, the brain reacts differently to cloned voices than to natural ones.
The brain processes cloned voices differently to natural voices, say researchers at the University of Zurich.
(Image: petrmalinak/Shutterstock.com)
The brain reacts differently to deepfake voices than to natural voices, although people often perceive cloned voices as natural. Depending on whether a natural voice or a deepfake is heard, different areas of the brain are active differently. Researchers at the University of Zurich have discovered this. They have identified two areas of the brain that react differently to acoustic signals.
According to the researchers, each person has a unique voice profile that helps to identify them. However, it is now possible to clone voices very realistically.
Participants should distinguish natural voices from clones
The team first checked how much human identity was retained in the deepfakes. The team recorded four male voices and copied them. The 25 test subjects were asked to compare the voices and identify whether the voices had the same identity. They were played two natural voices or one natural and one cloned voice.
The participants recognized the deepfake identities correctly in two thirds of the cases. "This makes it clear that although current deepfake voices do not perfectly imitate identity, they have the potential to deceive people's perception," says Claudia Roswandowitz, post-doctoral researcher at the Institute of Computational Linguistics.
Deepfakes seem to be less appealing to the brain
In the study of active brain areas, one part of the mesolimbic system made a difference in recognizing deepfake voices: the nucleus accumbens, an important component of the brain's reward system. When the participants heard two natural voices, this area of the brain was particularly active.
The auditory cortex, which analyzes sounds, also switched on in the participants. It was more active when the team played a deepfake and a natural voice. "We suspect that this area reacts to the imperfect acoustic imitation of the 'deepfake' voices and tries to compensate for the missing acoustic signal," explains Roswandowitz. The difference in activity was greatest when the deepfake voice sounded more unnatural and unappealing than the original. The researchers conclude that the fake part in the cloned voices reduces listening pleasure. "In particular, the neural mechanisms identified in the processing of deepfakes illustrate the human resistance to fake information, which we encounter more and more frequently in everyday life," says the scientist.
(are)