Saving ads and branding can be a rewarding job for celebrities and influencers. But is it too hard work? It is bet by the US company Veritone. Today, the company is launching a new platform called Marvel.AI which allows creators, media characters, and others to produce deeply counterfeit clones of their voices to license as they wish.
“People want to make these deals, but they don’t have enough time to go to the studio and produce content,” Veritone President Ryan Steelberg says Limit. “Digital Influencers, Athletes, Celebrities and Actors: This is a huge asset that belongs to their brand.”
With the help of Marvel.AI, he says that anyone can create a realistic copy of their vote and use it however they want. As Celebrity Y sleeps, their voices may be gone, recording radio points, reading audiobooks, and more. Steelberg says the platform will even be able to evoke the voices of the dead using training data from archiving recordings of AI models.
“Whoever owns the copyright to these sounds, we work with them to bring them to market,” he says. “It’s the job of the rights holder and they think it’s appropriate, but hypothetically, you can get Walter Cronkite to read the evening news again.”
Speech synthesis has improved rapidly in recent years, and machine learning techniques make it possible to create increasingly realistic voices. (Just think of the difference between how Apple’s Siri sounded when it was launched in 2011 and how it sounds now.) Many large technology companies, such as Amazon, offer ready-to-speak text-to-speech models that produce robotic but not unpleasant sounds. But new companies are also making boutique sound clones that sound like certain individuals, and the results can be almost distinguishable from the real thing. Just listen to this sound clone for example, podcaster Joe Rogan.
This advancement in quality motivates Veritone to create Marvel.AI, says Steelberg, and the potential of synthetic speech to match the company’s existing businesses.
Although Veritone markets itself as an “artificial intelligence company,” much of its revenue apparently comes from old-school advertising and content licensing. As Steelberg explains, its advertising subsidiary Veritone One has invested heavily in podcast space and invests more than 75,000 “ad integrations” with influencers each month. “It’s mostly native integrations, like product placements,” he says. “The ability to donate sponsorships and ads. It’s very effective, but very expensive and time consuming.”
Another part of the company, Veritone Licensing, licenses videos from several large archives. These include archives owned by broadcasters such as CBS and CNN, as well as sports organizations such as the NCAA and the US Open. “When you see Apollo moon landing material appearing in movies or the content of Tiger words in a Nike ad, all of this comes through Veritone,” Steelberg says. This experience in licensing and advertising gives Veritone an advantage over start-ups for artificial intelligence, focusing purely on technology, he says.
Marvel.AI provides its customers with two streams for customers. One is a self-service model where anyone can choose from a list of ready-made sounds and create a speech as needed. (That’s how Amazon, Microsoft, and others have been doing it for years.) But another stream – the “focus,” says Steelberg – is going to be a “controlled, white-glove approach,” where customers deliver training information, and Veritone creates a sound clone just for them. The resulting models are stored in Veritone systems and are available to produce sound at the time desired by the customer. Marvel.AI also acts as a marketplace, allowing potential buyers to send requests to use these votes. (How this is all priced is not yet clear.)
Steelberg is convinced that there is a demand for these votes and that Veritone’s business model is ready to go. But one important factor decides whether Marvel.AI: the quality of the artificial intelligence produced by the platform succeeds. And this is much less certain.
When asked about examples of the company’s work, Veritone shared three short clips Edge, each one-line entry of the mint brand. The first line reads Steelberg himself, the second AI clone, and the third sex-switched voice. You can listen to all three below:
The AI clone is at least in my ear, a pretty good imitation, though not a perfect copy. It’s smoother and cut than the real thing. But it’s also not a perfect indication of what votes can do during approval. Steelberg’s delivery lacks the enthusiasm and passion you’d expect from a real commercial (we don’t blame him for that – he’s neither a director nor a voice actor), so it’s not clear if Veritone’s sound models can get a full range of emotions.
Nor is it a great sign that the sound of the platform sizzle reel (embedded at the top of the story) was made by Steelberg himself and not a copy of the artificial intelligence. Either the company didn’t think the sound clone was good enough to work with, or it took time to make – either way, it’s not great product support.
However, technology is advancing rapidly, and Steelberg wants to emphasize that Veritone has the resources and expertise to embrace the new machine learning models that will emerge in the coming years. Where it distinguishes itself, he says, manages clients ’and clients’ experiences of the actual introduction of synthetic speech.
One problem that Steelberg offers is how synthetic speech can undermine the power of recommendations. After all, the appeal of a brand depends on the belief (regardless of misleading) that this or that celebrity really enjoys this particular grain / toothpaste / life insurance brand. If a celebrity can’t bother to express their approval themselves, won’t it remove the selling power of the ad?
Steelberg’s solution is to create an industry standard for publicity – some kind of voice that sounds before a synthetic speech a) to tell listeners that it’s not the right voice, and b) to reassure them that the owner of the voice supports that use. “It’s not just about avoiding the negative meanings of consumer cheating, but also wanting them to trust it [this or that celebrity] really accepted this synthetic content, ”he says.
Such issues are becoming increasingly important as synthetic content becomes more common, and it is clear that Veritone has been thinking hard about these things. Now the company just needs to convince influencers, athletes, actors, podcasters and celebrities to give it a voice.