ElevenLabs AI Founder Has an Concept for Preventing Audio Deepfakes

• AI voice tech firm ElevenLabs is grappling with deepfakes.
• The expertise, whereas progressive, has the potential for misuse, resulting in issues from lawmakers.
• ElevenLabs CEO stated digitally watermarking artificial voices is an answer.

Corporations on the forefront of AI voice expertise are grappling with how one can regulate deepfakes with out stifling innovation. 

“It should be a cat-and-mouse sport,” Mati Staniszewski, the cofounder and CEO of ElevenLabs, instructed The Atlantic. 

ElevenLabs —  which vaulted to a $1.1 billion valuation after launching in beta final yr — makes use of AI to generate convincing audio clips. That features text-to-speech voiceovers, dubbing audio into 29 languages, and cloning voices. The corporate claims its customers generated over 100 years of audio up to now yr. 

Nonetheless, lawmakers fear the expertise has a harmful potential for abuse.

Advances in AI have correlated with an increase in supercharged cellphone scams wherein imposter scammers solid themselves as love pursuits, relations, or authorities officers. Biden’s AI chief, Bruce Reed, has even stated “voice cloning” is the one factor that retains him up at evening.

And final yr 4chan customers exploited the device from ElevenLabs to generate deepfakes of celebrities spewing racist and transphobic content material, in response to Vice.

However Staniszewski is an idealist.

He sees ElevenLabs’ expertise contributing to a world the place sufferers with neurodegenerative illnesses like ALS can nonetheless talk of their voice after they lose the power to talk. It additionally has potential as a device to assist individuals talk throughout cultures and languages.

New York Mayor Eric Adams has been making robocalls in Mandarin, Yiddish, and Haitian Creole with ElevenLabs expertise and stated he is been in a position to attain extra of town’s non-English talking residents.

To capitalize on this potential whereas stopping fraud, Staniszweski stated customers ought to be capable to determine AI-generated voices from human ones. The “true answer,” Staniszewski instructed The Atlantic, is digitally watermarking artificial voices so people can differentiate actual from faux.

The corporate is creating the expertise, nevertheless it’ll solely be efficient with the cooperation of different firms. ElevenLabs has signed an accord with a number of different AI firms, together with OpenAI, Anthropic, Google, and Meta to fight deepfakes within the 2024 election.

ElevenLabs didn’t instantly reply to a request for remark from Enterprise Insider.