Brain reacts differently to deepfake voices than to natural ones

The human brain reacts differently to natural voices and so-called deepfake imitations. This has now been discovered by a research team at the University of Zurich (UHZ). His study shows that although people often mistake the stolen identity in a deepfake audio for a natural voice, their brain does not.

The latest algorithms for voice synthesis are now able to create deceptively real voice clones, according to a UZH report. This means that it has never been easier to imitate natural voices with deepfake technologies, either for fraud attempts on the phone or to give the voice assistant the voice of your favorite actress. But the brain is not so easily taken by surprise.

For the study, the researchers first recorded the voices of four male speakers and generated deepfake voices from them. In the main experiment, 25 test subjects were asked to find out whether the voices were artificial or natural. In two thirds of the cases, they were able to match the falsified votes. "This makes it clear that although current deepfake voices do not perfectly imitate identity, they have the potential to deceive people's perceptions," Claudia Roswandowitz is quoted as saying. She is a post-doctoral researcher at the Institute of Computational Linguistics at UZH and first author of the study.

At the same time, however, the researchers discovered previously unknown reactions in two areas of the brain. Imaging methods showed significantly less or more activity when processing deepfake audio than when listening to natural voices. Roswandowitz concludes from this that people "can therefore only be partially deceived by deepfakes". "In particular, the neuronal mechanisms identified in the processing of deepfakes illustrate the human resilience to fake information, which we encounter more and more frequently in everyday life." ce/js