FTC names winners in AI voice authentication challenge
Anderson Piza/Getty Images
Winners came from the private sector and academia as the agency continues its crackdown on AI-generated voice cloning.
The Federal Trade Commission announced winning submissions from its Voice Cloning Challenge, a federal initiative designed to help protect the public from artificially-generated voice cloning harms.
Three submissions were chosen to share $35,000 in prize money. These applications all propose solutions to differentiate genuine human voices from technologically generated voices transmitted digitally. A fourth submission was recognized but didn't share in prize money.
“Tapping American ingenuity is critical to solving big abuses like deceptive voice cloning,” Samuel Levine, the director of the FTC’s Bureau of Consumer Protection, said in a release issued on Monday announcing the awards. “When it comes to AI-driven fraud, the FTC will continue using every tool to deter harmful practices, shut down bad actors, and spur innovative proposals to help protect consumers."
Two of the winners came from private sector backgrounds while one came from academia. OmniSpeech, a software developer, submitted a tool leveraging artificial intelligence algorithms to detect subtle distinctions between real and synthetic speech and vocal patterns. OriginStory, another winner, submitted a similar application that uses sensors pre integrated into a bevy of software products.
OriginStory’s sensor technology aims to also authenticate which digital voices are from humans, measuring speech acoustics and co-occurring biosignals in the throat and mouth.
DeFake, from computer science professor Ning Zhang from Washington University in St. Louis, proposes watermarking and labeling voice recordings that have been created from pre-recording of voices sources from social media and other digital sources.
The proposed watermark would add distortions to voice samples imperceptible to the human ear, and make it harder to clone.
The fourth submission, from Pindrop Security, featured a voice cloning detection system that identifies audio deepfakes and voice clones in real time.
The FTC is one of the federal agencies spearheading action into policing AI-generated content and deepfakes, targeting robocalls and biometric softwares in recent decisions.