Veritone, a company based in the United States, launched a new platform called Marvel.ai, an AI that will generate deep fake voice clones of celebrities and influencers as they wish.
Veritone president Ryan Steelberg claims that with Marvel.AI, anyone can make a practical replica of their own voice and use it however they see fit. When celebrity Y is sleeping, their voice could be recording radio spots, listening to audiobooks, and more. Steelberg claims that the app would be able to restore the voices of the dead by training AI models with archive recordings.
“Whoever has the copyright to those voices, we will work with them to bring them to the marketplace,” he says. “That will be up to the rightsholder and what they feel is appropriate, but hypothetically, yes, you could have Walter Cronkite reading the nightly news again.”
In recent years, speech synthesis has advanced exponentially, with machine learning techniques allowing for the development of increasingly realistic voices. (Take, for example, the contrast between how Siri sounded when it first debuted in 2011 and how it sounds now.) Many large tech companies, such as Amazon, sell off-the-shelf text-to-speech models that produce robotic but not offensive voices at scale.
New firms, on the other hand, are creating boutique speech copies that sound like unique people, with effects that are almost indistinguishable from the real thing.
Even though Veritone describes itself as an “AI company,” it appears that traditional advertising and content licensing provide the majority of its sales. According to Steelberg, Veritone One, the company’s advertising subsidiary, is heavily involved in the podcast market, placing more than 75,000 “ad integrations” with influencers every month.
He states that the majority of the integrations are native, such as product placements. “It’s finding talent to voice advertisements and sponsorships. That is very effective, but it is also very costly and time-consuming.”
Veritone Licensing, another division of the company, licenses video from a variety of large archives. This includes the files of broadcasters such as CBS and CNN, as well as sporting associations such as the NCAA and the US Open.
“When you see Apollo moon landing footage in movies or Tiger terms material in a Nike commercial, it all comes from Veritone,” Steelberg explains. Veritone would have an advantage over AI startups that are solely focused on technology, he claims, because of its licensing and advertising expertise.
Steelberg persuades us that there is a need for these voices and that Veritone’s business model is sound. However, one big aspect will determine whether Marvel.AI succeeds: the quality of AI voices the site is capable of producing. And this is much less certain.