AI voice clones pose an ‘existential crisis’ for actors: ‘It’s a violation of our humanity’

This post was originally published on this site.

Nick Meyer said $100,000 would have changed his life.

The 26-year-old actor said it would have “taken a lot of weight” off his shoulders and provided relief for his family. Although he’s been acting professionally for a decade, Meyer said he makes less than $10,000 a year from acting and supplements his income with food service and retail jobs. So why would he turn down a voice-acting gig offering roughly 10 times his annual acting salary for only 20 hours of work?

Because the job entailed recording his voice to train artificial intelligence-powered voice replication models. “I am not going to sacrifice my morality for a paycheck, no matter how big,” Meyer said.

The L.A.-based performer is one of many voice actors reckoning with AI’s industry disruptions. Voice cloning has become much easier, requiring just seconds of audio. This poses a host of challenges for actors who have found their voices replicated online without their consent, knowledge or compensation, reducing paid job opportunities and stripping them of their agency.

When Meyer made it clear to his representatives in February that he was not going to take the gig, he said he was met with ire. He ended up parting ways with his agents after they told him they would not be a good fit going forward if he turned down the job. Meyer declined to name the agency, but The Times reviewed email exchanges between the actor and his former agents that verify the events.

About a year ago, Meyer said his voice was replicated without his permission by users of the popular AI chat platform Character.AI. Users cloned recordings of his voice and created online personas to accompany the voices. There are at least a dozen “Nick Meyer” characters featuring his name and image on the app, and they have collectively engaged in more than 100,000 chats — defined by the number of “human messages” sent to those characters. So Meyer knows what it’s like to not have control over what his voice is saying.

“If this gets any better, if this continues to get trained, if this has more footage or more recording of my voice, how much closer can it get to sounding like me?” Meyer said.

A Character.AI spokesperson said in a statement to The Times that the company takes “swift action to remove reported Characters that violate copyright law and our policies.” Meyer said he has reported the characters as unapproved uses of his name, likeness and voice.

During the course of reporting this story, Meyer’s cloned voice was replaced with generic voices, but the characters that bear his name and image haven’t been taken down.

“Users create hundreds of thousands of new Characters on the platform every day,” the statement continued. “Our dedicated Trust and Safety team moderates these Characters proactively and in response to user reports, including using industry-standard blocklists and custom blocklists that we regularly expand.”

Nearly a dozen actors interviewed by The Times said they are fearful of what their voices could be used for if they’re cloned without their knowledge. Whether that content is a violation of exclusivity clauses they signed with existing clients or something they morally disagree with, voice cloning could hurt more than just their wallets.

About 80% of working voice actors aren’t represented by a union, so the onus often falls on the individual to protect themselves. Up until a few years ago, worries about voice cloning were virtually nonexistent. Now, they concern thousands in the industry.

“It’s like the Wild West,” said Joe Gaudet, a Connecticut-based voice actor with more than 20 years of experience. Gaudet, 41, voiced more than 30 videos for a company before he says it replicated his voice and cut him out of additional work by using the clone for quick edits to scripts.

Gaudet said he was gutted, especially because he believed the company was working in good faith.

“You feel like you’re useless and you have no value,” he said. “It’s the worst feeling in the world. It’s the worst. And I know it’s not just me. These people in many, many companies are screwing people over.”

The National Assn. of Voice Actors aims to help performers navigate this essentially uncharted territory. The nonprofit, founded in March 2022 with the goal of providing healthcare for freelance voice actors, has become a crucial source of AI information and guidance for many in the industry. The organization crafted a contract rider that addresses many actors’ concerns about their voice being cloned or used to train AI models.

Although several actors said the rider’s language is now a non-negotiable part of new contracts, it doesn’t help those who signed contracts with expansive and vague language before the advent of AI. Agreements commonly include verbiage that actors’ recordings can be used in all “technology known or yet to be developed” or “in perpetuity throughout the universe.” Others have language buried in the fine print that enables companies to sell an actor’s voice to other parties.

The women behind the voices of Siri and TikTok speak out

Atlanta-based voice actor Susan Bennett is among the performers who signed vague contracts decades ago, not anticipating the advances in voice replication technology.

On Oct. 14, 2011, Apple released the iPhone 4s, which introduced the digital voice assistant Siri. Siri was, at the time, novel — she was the first interactive voice that didn’t sound robotic or monotone. And she was even programmed to have a bit of humor and sarcasm (in response to the question “What are you wearing?” Siri would say, “Aluminosilicate glass and stainless steel. Nice, huh?”).

Bennett received an email that day from a friend and fellow voice actor, asking if it was her voice.

“I went, ‘Well, gee, I don’t remember doing that work. I certainly didn’t get paid for that work,’” Bennett recalled. “It was a conflict of feelings, of course. I was very flattered that my voice was chosen, but on the other hand, it’s like, ‘Wow, there’s my voice, it’s just going to be completely ubiquitous, and how is that going to affect my livelihood as a voice actor?’ And, of course, there’s no way to really measure that.”

Six years before Siri’s launch, Bennett worked on a project with software company ScanSoft to create interactive voice recordings. She spent several months recording nonsensical phrases such as “Say bow geeky preface today” and “Say the doesn’t ding again” to capture as many sound variations as possible. After months of tedious voice-over work, she was paid by ScanSoft and sent on her way. She didn’t think about the project again until fall 2011, when her voice was suddenly everywhere.

Bennett, 75, said she knew her voice would be used for interactive text-to-speech technology, but she had no idea about the scale or reach. She said she wasn’t notified that she would be the voice of Siri or compensated by Apple. A representative for Apple did not respond to The Times’ requests for comment.

“I was extremely naive about what I was doing,” Bennett said. “It’s like, ‘Oh yeah, here I am, saying everything that could possibly be said. What could go wrong?’

“They could have thrown me a bone, sent me a few thousand and pat me on the head,” she said.

Years after Bennett’s debacle, Canadian voice actor Bev Standing found herself in a similar situation. TikTok debuted a text-to-speech generator in late 2020 that had a strong resemblance to Standing’s voice.

Standing’s first thought after friends and family sent her videos featuring her voice was, “What’s TikTok?” Standing had done recordings a few years earlier for a different company that said her voice would be used for Chinese translations.

When Standing saw a video that featured foul language in her voice, she knew similar problems would keep cropping up. TikTok’s text-to-speech feature has few content restrictions, so users could use Standing’s voice to say almost anything.

Standing said she wasn’t informed or paid by TikTok ahead of the release of the feature, so she sued its parent company, ByteDance, in 2021.

“You can’t do it to a movie star. They stand up and their lawyers stand up and their agents stand up. But when you’re a little nonunion person that lives in the middle of nowhere, no big deal,” Standing said. “Wrong. It’s a big deal. And because I spoke up, and because people took note, they’re standing up, and there’s a lot to be said in doing things in numbers.”

The complaint was settled out of court about four months after it was filed. Standing cannot discuss the terms of the settlement, and TikTok did not respond to The Times’ requests for comment.

The threat voice cloning poses is not limited to those with hours of high-quality recordings of their voices online. Realistic voice clones can be created with as little as three seconds of audio, said Tim Friedlander, president and co-founder of NAVA.

“If you have a video on social media somewhere that has your voice, image, name and likeness in it, it is in a system somewhere,” Friedlander said. “It has been used to train something, and it will more than likely be used to be sold back to you as a product in some capacity at some point.”

‘A violation of our humanity’

There’s a significant financial effect on actors when their voices are replicated and they are left to essentially compete for jobs with a cheaper version of their own voice. It’s often difficult for performers to track where their cloned voices end up or how they’re used, so it’s almost impossible to quantify the monetary impact of unauthorized clones.

Paul Skye Lehrman and Linnea Sage, New York City-based voice artists, discovered that both of their voices were cloned by AI company Lovo in 2022 and 2023. The married couple was listening to a podcast — ironically, about the dangers of AI — while driving when they recognized Lehrman’s voice, or rather, a clone of his voice. They estimate that their voices could have been used for “hundreds of thousands of scripts around the world.” Lehrman’s voice was the default option on Lovo for roughly two years, according to the complaint he filed last year in court. The company’s co-founder Tom Lee confirmed on the podcast “Category Visionaries” in 2023 that the technology had been used to create more than 7 million voice-overs at the time.

“Voice is as personal as our fingerprints,” Lehrman said. “It’s just such a violation of our humanity and an invasion of our privacy. It felt like being violated. And then everything — fear, anger, shame — all of this came with it.”

Sage and Lehrman worked with distinct clients on Fiverr, an online marketplace for creative freelancers, in 2019 and 2020, respectively. They now believe those clients were working for Lovo without disclosing their identities or motives. Both actors said they asked the clients — who had the anonymous usernames “User25199087” and “tomlsg” — in advance for the explicit purposes of the recordings they were submitting. They said they were told, unequivocally, that their voices would not be used for commercial purposes — only for research and internal purposes — without any mention of AI.

Lehrman and Sage claim that Lovo, without informing or paying them, cloned their voices and made them available for use on the site under fake names and for promotional materials. They sued Lovo in May 2024, and the case is ongoing. The company did not respond to requests for comment.

“We’re in a unique position to hold our destroyers accountable, and we are going to continue fighting the Goliath for everybody in our industry, to at least set some sort of message that you just cannot do this,” Sage said. “You can’t take advantage of actors and artists.”

Remie Michelle Clarke, an Irish voice actor and writer, came across her voice on the AI-powered narration site Revoicer, a company she’d never worked for. Clarke had booked a text-to-speech gig for Microsoft Azure in 2020, not understanding that the recordings could be used by third parties. She said the job description indicated that the recordings would be “mainly for internal use, and possibly for end use down the line.”

That possibility was more probable than she expected. When Clarke’s voice appeared on Revoicer in January 2023, the mom of two young children said she worried her voice would be used for nefarious purposes.

“My older boy, who’s nearly 3, is starting to hear my voice on the radio and TV and knows it’s Mummy. And I just wonder when he gets a bit older and he comes across things on the internet that might be very unsavory and hears Mummy’s voice — that makes it extremely personal and extremely difficult for me,” she said.

Clarke’s contract with Microsoft gave the company the rights to her voice recordings in perpetuity. A Revoicer representative declined to comment, but a developer confirmed to the Washington Post in 2023 that the company had a licensing agreement with Microsoft, which would have given it access to Clarke’s sample.

“The allusion to ‘The Little Mermaid’ has been used so many times, but this is it. It’s Ursula scraping the bottom of the ocean to try and get absolutely everything that they can at the expense of culture, at the expense of art, at the expense of individuals, families, societies,” Clarke said. “It’s huge, and it’s all taking far too long for it to change for the better.”

Clarke said her voice has since been removed from the site after she spoke about the situation in several interviews.

A glimmer of hope

Some actors are trying to embrace voice cloning to stay ahead of the curve. Bob Carter, a seasoned Atlanta-based voice actor and owner of recording space and voice-over education center the Neighborhood Studio, worked with AI company ElevenLabs to create a highly realistic clone of his voice. He’s paid every time his voice clone is used and can set parameters for how it’s utilized.

“I knew that there’s no stopping this. This train has already left the building. It is off and running,” Carter said. “I had to protect myself.”

Carter said the voice of his wife — actor and coach September Day Carter — was used without her knowledge, consent or compensation for a slew of projects.

“It’s always better to be proactive than reactive,” said Carter, 52. He’s now paid every eight days by ElevenLabs and said he takes comfort knowing he’s benefiting from how AI is transforming the industry, although he realizes some of his peers are hesitant to embrace the technology. “Change is scary when it happens to us, but it’s a good thing when it comes from us,” he said.

In addition to engaging voice actors directly, ElevenLabs has multiple safeguards in place to prevent users from cloning others’ voices.

“There is no single safety mitigation that is completely effective in preventing misuse on its own,” said Artemis Seaford, head of safety at ElevenLabs. “So what you want to have is essentially a safety stack, which is a series of safeguards that work together in order to provide a robust system against abuse.”

Some of those safeguards include a proprietary voice verification technology and several layers of screening and moderation to ensure users are using the technology only to clone their own voices.

A few states, including California and New York, are enacting legislation to protect against the misuse of unauthorized digital replicas, including video deepfakes and AI voice clones. But performers and creatives outside of those states remain at risk without federal legislation.

The Nurture Originals, Foster Art, and Keep Entertainment Safe Act (NO FAKES Act), introduced by U.S. Sen. Chris Coons (D-Del.), aims to address that gap. Scott Mortman, a lawyer and AI advisor who works with NAVA and teaches a course on AI law at Purdue University, said he’s “not optimistic” the law will pass anytime soon, despite its bipartisan support.

“Lord would hope if the two parties can agree on anything it would be the need to restrict the unlawful use of somebody’s image or voice or likeness, but that is to be determined because this administration overall appears to be quite resistant to any form of regulation and seems to be making a great effort to undo existing regulations,” Mortman said. “So whether or not this particular regulation ultimately gets signed into law very well may depend upon the person who has to sign it into law.”

As actors contend with quickly evolving voice replication technology and the threat of its misuse, many seem more aligned with Meyer, the 26-year-old who turned down a lucrative AI voice clone job, than Carter. Whether his voice would be distinguishable or just one of many voices layered to create a new product, Meyer said he didn’t want to be “complicit in the destruction of digital media.”

Meyer said those who deem voice cloning just the latest in a string of technological advancements in Hollywood, like CGI, are not seeing the full picture. CGI, he said, “made it easier to tell stories that were once thought impossible to tell,” solving a problem. To Meyer, voice cloning doesn’t come close to accomplishing that goal.

“It created a problem that didn’t exist.”