A.I. Has Hollywood Voice Actors Trapped Between Peril And Possibility

You've likely seen split-screen videos of actors wearing motion capture suits performing on a soundstage alongside the fully rendered image which appears in the final product. This AI technology for voiceovers works in a similar manner, only with audio.

"They'll hire an actor to basically be a puppeteer for the James Earl Jones voice," explains Keythe Farley, an actor and voiceover director who has credits on the Mass Effect games, "Adventure Time," and Cyberpunk 2077. "So the actor's performance will then get 'skinned' by James Earl Jones' [AI-generated voice]." And it's not just celebrities who provide those "skins," or customized voices that are overlaid onto a baseline performance.

Abigail Savage has appeared on shows like "Orange is the New Black" and "What We Do in the Shadows." She's uniquely positioned to talk about this technology, because she's not only an actor, but she's also been working as a sound editor and sound designer for over 20 years. She's worked closely with Respeecher, the same company that Lucasfilm teamed with for the new Darth Vader vocals, and was "smitten" with what the technology was able to do.

For a project Savage was working on that planned to utilize this technology, Respeecher had to recreate someone's voice, and in her estimation, it "seemed to have jumped over the uncanny valley into something that was quite usable and realistic and incredibly impressive." That particular project ended up not coming to fruition, but she was so taken by what the company was able to accomplish that she became a beta tester for its voice marketplace program.

"They have a bunch of pre-set AI voices they've developed [using] synthesized other human beings, so the voices aren't actual individuals, but are amalgamations of people. It's a way to record yourself saying something and choose whatever voices you want from their marketplace. Then it outputs those voices with what you said, and it maintains your performance but through another voice. That's something that's just great for a sound editor who's doing things and needs the occasional line said. It used to be you'd just spend a night doing group loop, which you'd get a bunch of friends and some pizza and you have people come into the booth and say the one random line that you need covered. Now, I can do it all myself, which is just, frankly, awesome."

Duncan Crabtree-Ireland is the National Executive Director of SAG-AFTRA, the labor union that represents actors across film, television, radio, and more. He's recognized an uptick in AI quality, but also doesn't think it has progressed to the point where AI could fully replace human actors. "I think any of us who've used text-to-speech technology know that it is a lot better than it used to be, but it's still quite limited," he points out. "If you imagine listening to the Harry Potter books with text-to-speech technology versus listening to the audiobooks of the Harry Potter books, which are full of artistry and really performed as much as they are read, those are two very different things. I can't tell you that it could never be the case in the future someday that AI technology couldn't attempt to replicate that performance quality. But it's definitely not there now."

Keythe Farley isn't fully buying it, either. "I don't think that AI is going to replace actors doing scene work anytime soon. They keep making these claims about how it can be emotionally viable, but I don't know how two AI actors play out a scene together in that regard."

While none of the people /Film spoke to believe voice actors will imminently be made obsolete, most acknowledged being scared or at least concerned about the implications of AI on their profession. "It's really terrifying how great the technology is getting, and I think it's super dangerous," Strong admits.

Tim Friedlander, a voice actor with credits on "Hunter x Hunter" and "One Punch Man," likens this new wave of AI technology to the long-term effects certain innovations had on the music industry.

"You look back at music, it's the same. 'Oh, well, synthesizers are going to make piano players obsolete, and loopers and samplers and digital interfaces audio is going to make performers obsolete.' And to a certain extent, we are now, 30 or 40 years later, where you can go to a performance and you'll have everything running off a Pro Tools rig and one singer [on stage] ... So it has eliminated an entire generation of performers, of musicians. Potentially, it's going to have the same impact here."

Even so, Hollywood has not raced to replace actors en masse — not yet, anyway. "The hope that we have, at least any kind of shining light, bright light in this, is that producers still want a human person to direct," Friedlander says. "They still, for the most part, don't want to replace human voice actors, if it comes down to the performance."

"It's all about how clever the voice actor will get with what they were born with and what tools will exist in the future," Nancy Cartwright writes. "Innovating rather than fighting change is a good perspective to keep in mind. I believe that the artist can and will always find ways to use technology to create, rather than simply allowing technology to dominate them and make their art form a thing of the past."

"By the way," she adds, "the derivation of 'animation' is 'a bestowing of life ... filled with breath ... make alive ... spirit' — something unique to the artist and not a machine."

While artificial intelligence may seem like an existential threat to the voiceover profession, several of the people /Film interviewed are trying to find a silver lining. In Nancy Cartwright's view, "[AI technology] could provide more opportunities for voice actors and make their performances more accessible to a wider audience by learning how to create artificial changes to their voices or adding nuances to their voices that they couldn't perform physically before — for example, speaking in a higher or lower octave or accent."

That accent example is currently being put into practice at a company called Veritone. Working with clients like NCAA, the Australian Open, The Masters golf tournament, CBS News, and hundreds more, Veritone uses AI in several different ways, but over the last year, they've become a significant player when it comes to synthetic voices. Co-founder and CEO Ryan Steelberg believes every voiceover artist should have their own versions of their synthetic voice created. Then, each artist can essentially become their own boss by overseeing an AI model of their voice which can produce more sessions in a day than a human could physically record — and speak in languages the human artist can't actually speak.

"Voice talent is not just the sound of their voice," Steelberg says. "Voice talent and voiceover is how effective they are in speaking. Right now, we can create great-sounding clones, but how the presentation of that voice is melded with and into a production is still a talent. Our position with voiceover talents is we embrace them. We have lots of voiceover talents as customers. We create synthetic voices, and we help manage those voices for them. They need to view this as a tool ... You take a great American English voiceover talent. Tomorrow, they can set up shop and offer their voice and their capabilities of, ironically, voicing their voice in multiple different languages. We can actually help bring a lot more business opportunity for voiceover talent by them investing and being willing to work with synthetic models."

"The Price is Right" host Drew Carey recently experimented with using an AI-generated model of his voice during a portion of an episode of his SiriusXM radio show, and after it didn't go over particularly well, he said he wouldn't be using it again. The technology may not be quite advanced enough to take over hosting duties for a radio show yet, but imagine an artist being tasked with recording a large number of hyper-local radio spots. ("You're listening to the Big Ape, 95.1, WAPE, Jacksonville's number one hit music station.") If that artist has a synthetic version of their voice (and if they reach an agreement with their employer), they could farm out those hundreds of recordings to the AI model and sit back while the check rolls in. The same thing applies for an artist hired to record audio versions of articles on websites, or to narrate audiobooks.

There are practical uses for this in the sports world, as well. One of Veritone's clients is soccer announcer Alan Smith. For the World Cup, the company used an audio clone of Smith's voice to do real-time announcing for hundreds of games that he wasn't able to actually voice himself. Using the same real-time data that powers box score updates, the clone of Smith's voice can read that data and do play-by-play commentary for a match the real man has never even seen.

A prediction Steelberg could be right about, though, is that AI-generated voices may soon spell the end of so-called "soundalikes," actors who make their living by impersonating famous voices in a variety of official capacities. "[Tobias Meister] is the name of the voiceover talent who's famous for doing Brad Pitt's voiceover work in Germany," Steelberg says. "Those days are going to be numbered here. Brad Pitt and Tom Hanks and these people, when their show and their scene is being exported, it's going to be [an AI] version of their voice that's going to be speaking in [another language]."

It's funny Steelberg mentioned Tom Hanks, because the "Toy Story" and "Cast Away" star has actually had a secret weapon in the voiceover industry for years. Jim Hanks, Tom's younger brother, has been acting since the early 1990s and has credits on projects like "Robot Chicken," "JAG," and "Dexter." But he's also arguably the most famous soundalike in the world, having doubled Tom's voice as Woody from the "Toy Story" franchise since 1995. When you hear Woody's voice in toys, theme parks, video games, and Disney on Ice, you're hearing Jim's voice, not Tom's.

Since famous Hollywood nice guy Tom Hanks probably wouldn't create an English-speaking AI version of his voice that would take away work from his brother, Jim himself may not have to worry about being replaced. However, he knows others in his field may not be so lucky. "I think the soundalike industry is definitely vulnerable," he says.

Candi Milo has also done some soundalike work. "Part of the contract for a film is that you will do pickups or reshoots, but mostly audio pickups, as the actor's time allows," she says. "I cannot tell you how many times I have gone on a set because so-and-so won't do their pickups, or it's too expensive to fly Angela Lansbury in, so they hire somebody like me to go in and do two lines of her. That's how easy it is to do."

When Lansbury was in London during the making of "Mr. Popper's Penguins," the studio called Milo onto the Fox lot in Los Angeles to perform a couple of lines. She was paid a one-time fee and does not get residuals on the project, because she's not credited as an extra voice. This has been a common practice in Hollywood for decades, but AI models could conceivably bring that to an end.

And pour one out for all of the John F. Kennedy voice impersonators who have previously gotten work in movies like "Transformers: Dark of the Moon" and "Forrest Gump"; today, a production could work with the former President's estate and generate the lines they need using his actual voice as a foundation.

"If you want to reproduce a piece of history in a film, like a 'Forrest Gump' kind of scenario, maybe using the real JFK and using this technology to use his voice print to say, 'I believe he said he had to go pee,' maybe that would've produced a more convincing moment in the film," Maurice LaMarche concedes.

But Jim Hanks' concerns extend far beyond soundalikes. He suspects a rise in AI might lead to lower wages for all voice actors. "I've been in this industry long enough to watch earnings drop, because every time anybody who actually creates content on the producer end of it finds a way to pay talent less. So if you're going to allow people to use AI based on your voice, I suspect they're going to immediately go, 'Well, you're not going into the studio, so why should we pay you as much?'"

Regarding the examples provided earlier about artists licensing their AI voice to read hundreds of articles or audiobooks, Hanks doesn't imagine the outcome will be quite as rosy as it seems at first glance. "I can't imagine that those people would go, 'You know what? We want Morgan Freeman to read this article instead of this [text-to-speech voice],' because why would they pay somebody? They don't have to at this point."

Even prominent voice actors have had close brushes with sketchy situations. "I was approached to do a voice thing for an NFT project and they wanted to basically have me say a million different lines, really random things, and then they would teach this computer program to say anything in my voice," Tara Strong recalls. "And I said, 'So theoretically, they could make it say I hate a specific human, or type of human, or race, or say really terrible words that I wouldn't say in my real world?' And they're like, 'No, there are failsafes.' But let's be real a minute. Right? You can't really guarantee that."

Similarly, Maurice LaMarche says he was approached by one of the top studios to perform a particular impression. "They wanted me to basically read a telephone book into the microphone, and see if they could then create an animated character using the voice of the celebrity, who I presume refused to do this, and see if they could create this character with my very decent impression layering over samples of his real voice. They offered me a large sum of money, and I've turned it down flat. Do I think that stopped it in its tracks? No, but I would not be able to look at whatever the product was later and go, 'I helped with that.' I can't be the person that helped with that. I can't say, 'Well, they're going to do it anyway, so I may as well take the money and run.' I can't do it."

Unfortunately, there are plenty of people who would take that job (or one like it), especially people who are attempting to gain a foothold in the industry. Things can quickly become complicated.

"What you have is a cadre of underemployed actors that would be happy to replace you," Candi Milo laments. "They would be happy to do it, because they call it their way in. 'I've got my foot in the door, I'm just going to do this.' And I went, 'That is your entire body in the door. Because once you do that, there's no going back.' ... Screen Actors Guild, as a union to protect us, is so big and they're so disparate that [they're providing] different needs. Maybe I've done [the first entry in a franchise] and the studio wants to do part two and they don't want to pay more money, and I say, 'Fine, you can't use me, then.' But somebody says, 'Well, I want that job. They don't want Candi, and I want that job, and I can sound like her real voice, so I'm just going to do it.' Do they represent her, or do they represent me saying, 'I don't give permission for you to do this'?"

That constant fear of being replaced, relatable in so many professions operating in capitalist societies, is about more than just money for these performers. There's something indescribable that happens when an actor truly inhabits a role — something that can't be quantified, but can be felt by those who connect with the final result.

"There are definitely many times, particularly in voiceover, where the industry doesn't like to make you feel cherished as much as the fans do," Tara Strong says. "The [production] will say, 'Oh, we made this billion in toys' — which we don't have a backend on — 'and this billion on the movie, and we want to make the next movie and we want to give you guys scale and a half.' So you guys are making billions and you want to give us $2,000? [They'll say], 'Right, and if you don't do it, we're going to hire someone to copy your voice.' That doesn't feel so good. I don't know any actor that doesn't, especially in my world that I'm in every day, which is the animation world, put their heart and soul into something and feel very connected to a character. It's hurtful when they say, 'We could get someone else.' Because you really feel like once you are established as that character, it's a part of you. There's a part of me in everything that I do. That negotiation is a little soul-crushing. So I don't know that some genius company working with a Hollywood mogul would really give a s*** about honoring what a voice actor's created if they can mimic it."

Duncan Crabtree-Ireland specifically addressed a question many people had about the James Earl Jones scenario: Why didn't Disney just hire a replacement actor instead?

"That's a question that's impossible to answer, but what we can say is that so long as that work is being compensated and paid for, then the incentive to do it because of that is removed," he says. "So the company isn't choosing to do that because they're saving money. The company is choosing to do that because they believe they're getting the best creative output by using that iconic voice. And that is inherently a protection for our members, because if it's not being done for economic reasons, if it's being done for creative reasons and it's being done with consent and compensation, that preserves the whole range of other job opportunities that aren't tied to one particular performer's voice for everyone else. So really, I think establishing that principle that synthesized voices and AI voices don't translate into free work, that's the most important principle to help make sure our members are protected."

/Film reached out to multiple people at Skywalker Sound, but a senior publicist told us they are not able to participate in this story.

"I think that the misconception is that we are trying to stop this technology from coming into being, which is a fool's errand," Keythe Farley clarifies. "What we need to understand is that we are partners in this technology, and what the creators of this technology need to understand is that there are rules of the road, and that those rules need to be followed."

The James Earl Jones news conjured nightmarish imaginary scenarios involving voice actors. Consider a situation in which 20th Century TV decides it wants to keep "The Simpsons" running, but when it comes time to renew Nancy Cartwright's contract, the network decides that instead of hiring her, it would rather tap into her old archival recordings and use AI to generate new line readings for Bart Simpson for future seasons. Thankfully, that won't actually happen, and Duncan Crabtree-Ireland explained why.

"All of those episodes of 'The Simpsons' that have ever been recorded since it started, and everything going back decades before that, is subject to a reuse of footage of photography and soundtrack provision. It's in our collective bargaining agreement. It basically says if they're going to make any kind of other use of the soundtrack or photography of any project, they have to negotiate with the performers who are involved and pay compensation as part of that negotiation process. So in that example that you gave, if they wanted to use that as the basis for creating an AI voice model and then using that for further projects, they would have to negotiate with her for permission to do that. And if they did it without negotiating or getting permission, we, the union, can pursue that through an arbitration. There will be circumstances in which she could sue them in court if she wanted to. So there are legal remedies that are already built into the collective bargaining agreement to address any kind of use beyond the original use. And that includes making it part of an AI training data set, that includes use of clips, it includes any aspect of that kind of use."

While there are safeguards in place against major studios engaging in bad behavior, those don't apply to the comparative Wild West of deepfakes. But SAG-AFTRA is trying to protect its members from that, as well. Crabtree-Ireland explained that the union has had success in getting some legislation enacted in New York and California to help put those protections in place, but he would still like to see action taken at the federal and even global levels.

That's all well and good for union members. But what about people who aren't union members? In addition to downloading NAVA's free AI rider and applying that to their contracts, is there anything that can be done for them? Crabtree-Ireland said even non-union work can be done under union contracts that help protect people from being exploited.

"It is really hard to envision how you can adequately protect people from this sort of thing without using collective bargaining, because there is such a tendency for overreach on the part of companies who do this stuff. And if you aren't part of a collective bargaining group and you don't have a lot of leverage, it can be very hard to push back on demands that you do things like sign away your rights to your image, likeness, or voice. What I would just say to everyone who will be reading this is it is essential that you not do that. Once you've signed away, in perpetuity, the rights to your image, likeness, or voice, it is virtually impossible to get it back. We've seen many, many examples of people that have really been abused in this way. I know it's really hard to say no to that sort of thing, but it is essential that you say no to that thing. One way you can do that is simply by reaching out to us to help get a project organized. Even if you're not a member yet, we can help organize projects. And once they're under a collective bargaining agreement, now it's no longer the individual performer's job to have to fight that."

It's also worth noting SAG-AFTRA is working with other partners inside and outside the industry to, as Crabtree-Ireland put it, "establish a common set of principles for the human-centered implementation of AI technology."

"This goes beyond voiceover or even beyond just the entertainment industry," he says. "This is a broader issue."

If the past few months are any indication, artificial intelligence is going to be a major part of our lives going forward. That may be an uneasy realization for many, but Abigail Savage suggests we shouldn't be afraid of it.

"I think right now, people are uncomfortable with it unless it's very [clearly] specified in a project, 'This is an AI-recreated voice of this past person. They never really said these lines into a recording.' I think if the common audience knew the degree to which we are already manipulating reality when we are doing sound editing work, I think it would be less controversial. At some point in the future, it's probably going to be more normalized, because we're already cobbling together interviews to make people say sentences that were never actually spoken. I do documentaries all the time, and the degree to which we are imposing, manufacturing, a reality so that the director's viewpoint and the interviewee's viewpoint comes across in a more obvious or clear manner, I think would really shock most audiences ... At some point it will, I think, become a normal tool in our toolkit for being able to tell stories."

"CGI sucked when it first came out too, but it was abused everywhere," she concludes. "I think it took something away from filmmaking for a few years while we grappled with how much to use, how to use it, and the fact that there was a deep uncanny valley going on. Technology gets absorbed by artistry, and artistry eventually wins out."