Computer scientists at the McKelvey School of Engineering at Washington University in St. Louis prevent synthesis of deceptive speech used by scammers and attackers by making it more difficult for AI tools to read voice recordings.

“I’m very excited about our capability in fighting misuse of generative-AI capabilities,” said Ning Zhang, PhD, assistant professor of computer science and engineering.

Zhang is taking a stand against the misuse of deepfakes, synthesized speech that can be misused to deceive people and machines. So Zhang’s lab developed AntiFake to fight the growing problem. Zhang and first author Zhiyuan Yu, a graduate student, built the AntiFake tool to be generalizable and tested it against five state-of-the-art speech synthesizers. AntiFake achieved a protection rate of over 95%, even against unseen commercial synthesizers.

“We want to be able to fully protect voice recordings,” Zhang said. “It would be fully embedded so it automatically protects your voice.”