The AI craze is in full swing, and that includes several new companies that want to help artists and creators license their AI voice for commercial use. A new UK company called Myvox recently joined a list of companies betting on AI voice generation as a meaningful source of revenue.
Let's take a look at a just what AI voice generation is and a few different companies that aim to make it a legitimate business.
What Is An AI Voice And How Do They Come About?
When we talk about "AI voice generation," we're talking about an audio model that replicates what somebody would sound like if they were saying — or singing — whatever text a user puts into a prompt. If you've heard of "deep fakes," (videos of celebrities or other people that look real but are entirely computer generated), well, these voice interpretations are the audio version of that.
In other words, an AI voice sounds like a famous person, but whatever it says is completely up to a user inputting text. In simple terms, somebody creates a voice model based off existing audio files of a person's voice (it could be songs, speeches, conversations, interviews — whatever). When the model has enough examples of how a person may say or sing a certain phrase, it's eventually able to "replicate" their voice.
Originally, voice models needed a lot of different examples to create a believable replication. That's why you may have seen a lot of videos on TikTok etc. using the most famous voices around — Frank Sinatra, U.S. presidents, Taylor Swift, and so on. Now, these models start to need fewer audio samples to get closer to a believable voice.
Which is why even lesser-known artists or just everyday people may have their AI voice floating around out there somewhere.
Woah, This Sounds Like It Could Be Problematic
Oh, it definitely can be. As if there isn't already enough concern about misinformation out there, bad actors can easily create audio clips that sound like people saying things they never said. You can imagine the implications this may have in a landscape where a lot of people "share" first and verify second (or let's be honest, never verify).
That's not to say the use of voice deep fakes is only nefarious. There are some instances in which it can actually be pretty interesting. Consider educational environments, where people could "hear" historical events happen when a microphone wasn't around. Or even potential therapy applications. But the important part is that in order for these fake voices to be a force for good, we need some level of consent or ethical guidelines.
Beyond the ethical issues, AI voice generation presents a pretty serious legal dilemma. After all, somebody's voice is part of their own name, image, and likeness. It wouldn't be hard for somebody to, say, create a service that allows you to sell "shoutouts" from famous celebrities without their consent. We see this already in practice with the number of fans and accounts creating music with an artist's AI voice. Using AI likeness is also at the heart of SAG-AFTRA's strike against Hollywood studios.
And that's where some of these new startup companies come into play.
Startup Companies Trying To Help You License Your AI Voice
Indie/Alt Pop singer Grimes was one of the first artists to full embrace the concept of letting fans or other artists use her AI voice in new songs — provided she gets a cut of the money it makes. She partnered with a company called CreateSafe to make her own "Elf.Tech" voice model, which anybody can use to add AI Grimes to their tracks. She's also requiring people to use music distribute Tunecore to make sure she gets money from the licensed AI.
A new UK-based startup called Myvox (currently in beta) is now opening up that process to, well, people who aren't Grimes. The company's current voice model is dubbed Dahlia, but there's an application form for artists who want to use their voices that they can then license for a fee. Artists pay $9.99 per month to use it. If somebody wants to use Myvox's AI voice generating features commercially, they need to pay for the "pro" subscription (currently $14.99 per month), which allows them to distribute two songs per month using the model and also covers royalty payments from the master recording (again, after a 50/50 split).
Other companies doing similar things include DJ Fresh's startup Voice-swap (which is geared toward DJs and producers who don't want to use their own voice on tracks) and Holly Herndon's Holly+, which is her own AI voice model.
Also of note, Universal Music Group is reportedly talking to Google about licensing an artist's AI voice. This, of course, introduces a whole new swath of questions around the legality of licensing a creator's voice.
Are AI Voice Models Really A Money-Making Opportunity For Creators?
Possibly. It really depends on the application and each individual artist's fan base. There probably isn't that big of a market for, say, deep-faked music as a whole. After all, much of music fandom revolves around the concept of authenticity and story. Listening to what a modern-day Nirvana track "might" sound like is an interesting exercise, but it's probably not a business in-and-of-itself.
There's definitely an argument to be made for the utility of including AI voices on tracks when the music is ultimately meant to perform a function, rather than be story-building art. Think "focus" or "workout" playlists where it's more about the vibe and beat than the artist. It's like hanging a print on your wall of something that looks like it came from the brain of Andy Warhol. It adds some color to the room and takes up space. But it doesn't provide the same story that an Andy Warhol print might (or, you know, an original Andy Warhol).
However, there's no doubt that if a track features an AI voice of a living person, that person deserves compensation. Any company stepping in to make sure creators are protected are welcome, especially in an age where technology is moving quickly and not always with ethical interests in mind.