Whenever deepfakes are in the news, it’s almost always for the last ones terrifying way Bad actors have figured out how to spawn hoaxes or cyberbullies using AI-based technology. However, the media industry has found more practical (and less sinister) applications, such as using face swap to create more realistic visual effects, synchronize the actors’ mouths with the dialogue in dubbed movies, and now automate voice work.
Veritone, the creator of the world’s first operating system for artificial intelligence, aiWare, launched a new platform this week called Marvel.ai which allows content creators, celebrities and others to generate audio deepfakes of their licensed voices at their leisure.
Veritone’s rhetoric is that businesses and media personalities can produce monetized audio content and generate revenue without ever setting foot in a studio, because after all, time is money. Built on its AiWare operating system, Marvel.ai produces synthetic voice clips that he says sound like the real thing for radio spots, audiobooks, voiceovers, and localized content, among other examples.
“With full control over their voice and use, any influencer, personality, or celebrity can literally be in more than one place at once,” the company said in A press release Friday. “This would open the door to a new level of scale that was not humanly possible before, allowing them to increase the number of projects, sponsorships and supports they can achieve in any given year.”
The process, also known as voice cloning, uses artificial intelligence and machine learning algorithms to reproduce someone’s voice based on a series of audio samples. The Marvel.ai platform itself functions both as a marketplace in which customers can submit requests to use a particular voice model for auto-generated content, and as a self-service tool that looks like your more traditional text-to-speech player, allowing users to choose from a catalog of pre-generated voices to create a custom audio clip.
In an interview with the edge, Veritone chairman Ryan Steelberg has said he is offering another service – a “managed white glove approach” – where customers can submit voice clips to train Veritone’s systems and produce their own voice clone for licensing.
Speaking to the point of sale, Steelberg explained that while Veritone markets itself primarily as an AI developer, it also relies on old-fashioned advertising and content licensing for a large chunk of its revenue. Its advertising affiliate, Veritone One, is heavily invested in the podcasting industry and places more than 75,000 “ad integrations” with influencers each month, he said.
“It’s mostly native integrations, like product placements,” Steelberg told The Verge. “It allows talent to be heard in sponsorship and advertising. It is extremely efficient but very expensive and time consuming. “
This advertising expertise as well as technological advances in speech synthesis in recent years have motivated the company to build a better solution. Of course, taking off from Veritone’s platform largely depends on the conviction of its voice clones. Steelberg shared a few examples of Marvel.ai’s finished product with The Verge, which you can check out. here, and the clone looks human enough to me. The tone is a little odd in a way that I can’t quite put my finger on. However, I’ll be honest it’s hard to tell if it’s just my brain convincing it sounds to me because I know it comes from a robot and not from a person.
While I know I mentioned earlier that the media industry finds less sinister applications for deepfake technology, there is one aspect of Marvel.ai that gives me serious pause. Whoever owns the legal rights to one voice – that is, not just the person behind the voice – can use the platform to create the audio message they want. And that raises many of the same privacy red flags and misuse risks that have hampered the rise of deepfake videos. Marvel.ai can even resurrect the voices of the dead using archived recordings to train its AI systems, Steelberg told The Verge.
“Whoever owns the copyright in these voices, we will work with them to bring them to market,” he said. “It will be up to who owns the rights and what they think is appropriate, but hypothetically, yes, you could ask Walter Cronkite to reread the evening news.”
From a technological point of view, this is obviously impressive. Morally, it’s scary as hell. I really don’t need to hear Cronkite and other dead big names revived thanks to AI, especially since you know It’s only a matter of time before brands start doing shit, like using Princess Diana to endorse Oreos or David Bowie to get you to subscribe to Spotify Premium. I mean, did we learn nothing of Prince fiasco hologram?