Abbey Veffer has been working as a voice actor for three years. She’s early in her career. But she’s visible enough that in February, she got doxxed on Twitter.
Someone had created a fake account that looked just like her own, put her address on there, and then went a step farther.
“They had used an AI program to try and synthesize my voice,” she explains. “And it wasn't that bad of a likeness. … To an untrained ear, it was pretty close.”
But what was worse — “her” voice was saying explicit, racist things. For Veffer, it was like an out-of-body experience — online for all to hear, an evil twin turned loose on Twitter.
She was worried. Would her boss hear? Would she lose her job? More troubling, this had happened to three friends of hers, also voice actors.
“This kind of defamation is really harmful for people who are just starting out in their career, or even for the more-established,” she says. “It's just really upsetting.”
Eventually, Veffer got the tweet taken down (it took three months, owing to … well, Twitter), and no career damage done that she can point to. And though she never figured out who posted it, she certainly knows how it was made.
There are a bunch of new AI companies popping up that allow users to upload a few minutes of recorded audio of a voice and then clone that voice, getting it to say anything at all. The imitation isn’t perfect, but safe to say it’ll soon be indistinguishable from human voices. Anyone who has enough audio out there is a target. And voice actors produce a lot of audio.
Voice work is everywhere: Cartoons, commercials and video games, sure, but also audiobooks and corporate-training videos. And somebody has to lend a voice to the bus that tells you what stop you’re at.
Since ChatGPT came on the scene late last year, we’ve heard a lot about AI and all the good and bad things it might do for us. And we’re going to hear a lot more from AI, as it becomes easier, and cheaper, for businesses to have the robots do the talking.
Which raises some important questions, like, who should control this technology? Will it make the actors themselves obsolete? And when it’s used against someone, who’s responsible?
That one is a question Susan Eisenberg has had to grapple with. Eisenberg is a voice-acting legend: the voice of Wonder Woman in the cartoons “Justice League” and “Justice League Unlimited,” she also voices a lot of video games, including “The Elder Scrolls V: Skyrim.”
Like a lot of pop culture properties, this game has an avid fanbase. And the company behind the game, Bethesda Games, lets fans use its data to create new stuff — weapons or outfits or scenarios. These are called mods, and they’re fan fiction, basically. All of which is hosted on a site called Nexus.
In May, a fan wrote to Eisenberg to tell her some of the characters she played — including her voice – were used to create pornographic scenes. The results range from embarrassing to gross, and while the voice doesn’t sound the way Eisenberg talks, it definitely sounds like her.
“It was beyond upsetting, and I was reluctant to even look at what was going on, but I had to because that's my responsibility to myself — to see what's being used,” she says. “It's very unclear to me how to move forward with this, because who owns my voice?”
Bethesda Games did not respond to a request for comment. Nexus, the company that hosts all these mods, pointed me to its policy saying it will take down offensive content if the creator, or a voice actor, asks for it:
“... if we receive a credible complaint from a party who feels your work is damaging to them — this could be the voice actor or the body which owns the rights to the character/assets — we will remove the content if appropriate.”
In other words, Eisenberg and other actors will have to be vigilant, scouring the internet for clones, since any policing of their likenesses is up to them. And for now, legally, this is a gray area.
What Veffer and Eisenberg experienced is one of the big fears voice actors have about AI: having your voice cloned and used against you. The other fear comes straight out of old sci-fi stories: being replaced by the machines entirely.
To dig into that, let’s go to a recording studio behind a duplex off Melrose, where an actor named Matthew Parham is in the booth, trying to coax some magic out of the human voice box for a commercial.
I can’t tell you exactly what Parham is recording because companies are super secretive — video games, cartoons, whatever. In this case, I can tell you it’s a commercial for a product that’s best known as … orange.
Parham started his voice acting career sort of accidentally, during the pandemic. He does a lot of commercial work now, he says, “but mostly it's … brands that kind of need that ‘cool sound’ or ‘urban’ vocal print, is the dog whistle term for it.”
Brands sometimes fall over themselves trying not to sound racist, which draws attention to the fact that they’re asking somebody to sound Black without asking them to “sound Black.”
Parham and his director, Tim Friedlander (who owns this studio), have a kind of imaginary dial of Black voices. On one end, rapper Tupac Shakur; on the other, nonthreatening “Fresh Prince of Bel-Air” character Carlton Banks.
“So I just kind of figure out what the brand is, and go from there,” he says. “An ‘urban’ read for a tech company is not going to be on the Tupac side of things, they're kind of going to want more Carlton in the voice.”
Tim Friedlander directs Matthew Parham in the recording booth. Photo by Brandon R. Reynolds.
Voice work is an art and a business — tuning up certain attributes, cadences, accents, and moods, tuning down others. Successful voice actors are good at tuning in what the client wants, or needs, in an instant, whether that’s anime monsters, corporate training videos, or the voice of a city bus.
But what if a robot can do that? Someone can turn some actual dials on an app and bring up more Carlton with a hint of Tupac on a synthesized voice?
That presents an existential threat to a voice actor — just like AI writing does to those Writers Guild screenwriters striking now.
Friedlander, the director working with Parham in the booth, is addressing that threat directly. He’s a voice actor, too, and the founder of a nonprofit called the National Association of Voice Actors (NAVA). He started it as a way of getting health insurance for voice actors, but last year got pulled into the big conversation of our time: artificial intelligence. Specifically, what happens when AI can clone or create voices on demand?
“We started basically figuring out what we could do in that case — what can we do to help protect the voice actors?” he says. “What can we do to help protect somebody from having their voice taken and used without their permission, without their consent?”
Decorative pillows, including one showing actor Nicolas Cage’s face, are on a couch at Tim Friedlander’s recording studio. Photo by Brandon R. Reynolds.
This is a big part of a potential strike of SAG-Aftra members: making sure the studios don’t go full robot.
“In almost every discussion we've had about the strike, it comes up. … There's a possibility that somebody could step in and say, ‘See, we don't need the actors, we don't need the writers, we don't need the voice actor, because technology is good enough as it is,’” he says. “So you're gonna put a ton of people out of work … for what? To save some money, is what it seems like.”
SAG-AFTRA is certainly thinking about this as it continues to negotiate with the studios and streamers. The concern now is whether an actor owns their image, or voice, and if companies should be able to clone someone's voice or image, and how the actor would be paid for it.
“You need to have the person's informed consent about what is going to be done with voice models or other data. That's really their personal data that's being taken from part of making whatever project it is,” says Duncan-Crabtree Ireland, the SAG-AFTRA’s national executive director and chief negotiator. “And they need to be fairly compensated for it because it is their voice that's now being used as the voice of whatever project is going forward.”
The union will look out for union actors’ contracts, but according to a survey NAVA conducted, most actors are nonunion. (To that end, NAVA created a rider, which actors can try to insert into their contracts and prohibits the use of their voice data for chumming the AI waters.)
But voice actors are also excited about what AI can do for them. They can use their own clones to fill in gaps when they can't record, or when their voice is blown out — a vocal mini-me that does all the screams and tiny little edits that nobody wants to spend the time climbing into the recording booth for.
And actors could license their voices to be used for, say, creating voices for people with speech disabilities, or who are nonverbal, or as the voice that automatically turns written text to audio.
Or even doing boring stuff, like Friedlander did at the start of his career.
“I spent six months doing all the training manuals for GE transportation services on how to repair diesel engines for trains,” he says. “I spent months and months reading, ‘Three-quarters of a turn with the 5/16th wrench going into whatever hose with torque value this and this and this.’ That's something that's going to get updated, it's going to need to be changed … and if I have a synthesizer or synthetic voice that this company is authorized to use, they can compensate me and pay me for the licensing of my voice.”
It’s a brave, loud new world of clones doing a lot of work no human has the lung capacity for. I asked Parham if he’d license his voice.
“I definitely would lend my voice to something synthetic for the right price, with the understanding that it's going to be used with my consent,” he says.
Out of strictly professional interest, I had to know the number.
“Enough to where I don't have to work anymore — which, you know, inflation, I don't know, a couple million or something like that.”
Good to know, good to know (gonna write that down).
You could say voice actors are the canaries in the coal mine for how AI will change our work … and our online identities. And when they say they’re worried, or excited, maybe we should listen.