Companies at the forefront of AI voice technology are grappling with how to regulate deepfakes without stifling innovation.
“It’s going to be a cat-and-mouse game,” Mati Staniszewski, the cofounder and CEO of ElevenLabs, told The Atlantic.
ElevenLabs — which vaulted to a $1.1 billion valuation after launching in beta last year — uses AI to generate convincing audio clips. That includes text-to-speech voiceovers, dubbing audio into 29 languages, and cloning voices. The company claims its users generated over 100 years of audio in the past year.
However, lawmakers worry the technology has a dangerous potential for abuse.
Advances in AI have correlated with a rise in supercharged phone scams in which imposter scammers cast themselves as love interests, family members, or government officials. Biden’s AI chief, Bruce Reed, has even said “voice cloning” is the one thing that keeps him up at night.
And last year 4chan users exploited the tool from ElevenLabs to generate deepfakes of celebrities spewing racist and transphobic content, according to Vice.
But Staniszewski is an idealist.
He sees ElevenLabs’ technology contributing to a world where patients with neurodegenerative diseases like ALS can still communicate in their voice after they lose the ability to speak. It also has potential as a tool to help people communicate across cultures and languages.
New York Mayor Eric Adams has been making robocalls in Mandarin, Yiddish, and Haitian Creole with ElevenLabs technology and said he’s been able to reach more of the city’s non-English speaking residents.
To capitalize on this potential while preventing fraud, Staniszweski said users should be able to identify AI-generated voices from human ones. The “true solution,” Staniszewski told The Atlantic, is digitally watermarking synthetic voices so humans can differentiate real from fake.
The company is developing the technology, but it’ll only be effective with the cooperation of other companies. ElevenLabs has signed an accord with several other AI companies, including OpenAI, Anthropic, Google, and Meta to combat deepfakes in the 2024 election.
ElevenLabs did not immediately respond to a request for comment from Business Insider.
