Last summer, on their way to a doctor's appointment near their home in Manhattan, Paul Skye Lehrman and Linnea Sage listened to a podcast about the rise of artificial intelligence and the threat it poses to the livelihoods of writers, actors, and other entertainment professionals.
The topic was especially important for young newlyweds. They made a living as voice actors, and artificial intelligence technologies began to generate voices that sounded real.
But the podcast took an unexpected turn. To highlight the threat posed by artificial intelligence, the host conducted a long interview with a talking chatbot named Poe. He looked just like Mr. Lehrman.
“He was interviewing my voice about the dangers of artificial intelligence and the damage it could have on the entertainment industry,” Lehrman said. “We stopped the car and sat there, in absolute disbelief, trying to figure out what had just happened and what we should do.”
Mr. Lehrman and Ms. Sage are now suing the company that created the bot's voice. They claim that Lovo, a Berkeley, California start-up, illegally used recordings of their voices to create technology that competes with their voice work. After listening to a clone of Mr. Lehrman's voice on the podcast, the couple discovered that Lovo had also created a clone of Ms. Sage's voice.
The pair join a growing number of artists, publishers, computer programmers, and other creators who are suing makers of artificial intelligence technologies, alleging that these companies used their work without permission to create tools that could ultimately replace them in the labor market. (The New York Times sued two of the companies, OpenAI and its partner, Microsoft, in December, accusing them of using its copyrighted news articles to build their online chatbots.)
In their lawsuit, filed Thursday in Manhattan federal court, the pair alleged that anonymous Lovo employees paid them for some voice clips in 2019 and 2020 without disclosing how the clips would be used.
They say Lovo, founded in 2019, is violating federal trademark law and several state privacy laws by promoting clones of their entries. The lawsuit seeks class-action status, with Mr. Lehrman and Ms. Sage inviting other voice actors to join it.
“We don't know how many other people were affected,” said their lawyer, Steve Cohen.
Lovo denies the lawsuit's claims, said David Case, an attorney representing the company. He added that if all the people who provided voice recordings to Lovo gave their consent, “then there is no problem.”
Tom Lee, the company's CEO, said in a podcast episode last year that Lovo now offers a revenue-sharing program that allows voice actors to help the company create voice clones of themselves and receive a share of the money they earn from those clones.
The lawsuit appears to be the first of its kind, said Jeffrey Bennett, general counsel for SAG-AFTRA, the union representing 160,000 media professionals worldwide.
“This lawsuit will show people – especially tech companies – that there are rights that exist in your voice, that there is a whole group of people out there who make a living using their voice,” he said.
In 2019, Mr. Lehrman and Ms. Sage promoted themselves as voice actors on Fiverr, a website where freelance professionals can advertise their work. Through this online marketplace, they were often asked to provide voice work for commercials, radio ads, online videos, video games, and other media.
That year, Ms. Sage was contacted by an anonymous person who paid her $400 to record several radio scripts and explained that the recordings would not be used for public purposes, according to correspondence cited by the indictment.
“These are test scripts for radio commercials,” the anonymous person said, according to the lawsuit. “They will not be disclosed externally and will only be consumed internally, so they will not require rights of any kind.”
Seven months later, another unidentified person contacted Mr. Lehrman about a similar job. Mr. Lehrman, who also works as a television and film actor, asked how the clips would be used. The person said multiple times that they would only be used for research and academic purposes, according to correspondence cited in the lawsuit. Mr. Lehrman was paid $1,200. (He provided longer recordings than those provided by Ms. Sage.)
In April 2022, Mr. Lehrman discovered a YouTube video about the war in Ukraine narrated by a voice that sounded like his.
“It is my voice speaking about weapons in the Ukrainian-Russian conflict,” he said. “I turn ghost white: I have goosebumps on my arms. I knew I had never said those words in that order.
For months, he and Ms. Sage struggled to understand what had happened. They hired a lawyer to help them track down who had made the YouTube video and how Mr. Lehrman's voice had been recreated. But the owner of the YouTube channel appeared to reside in Indonesia and there was no way to find him.
Then they listened to the podcast on the way to the doctor's office. Through the “Deadline Strike Talk” podcast, they were able to identify the source of Mr. Lehrman's voice clone. A professor at the Massachusetts Institute of Technology had put together the chatbot using Lovo's text-to-speech technology.
Ms. Sage also found a video online in which the company had presented its voice technology to investors at an event in Berkeley in early 2020. In the video, a Lovo executive showed a synthetic version of Ms. Sage's voice and the compared with a recording of his real voice. They both starred alongside a photo of a woman who wasn't her.
“I was in their video presentation to raise money,” Ms. Sage said. The company has since raised more than $7 million and has more than two million customers worldwide.
Mr. Lehrman and Ms. Sage also discovered that Lovo was promoting voice clones of her and Mr. Lehrman on its website. After sending the company a cease-and-desist letter, the company said it had removed its voice clones from the site. But Mr. Lehrman and Ms. Sage argued that the software that handled these voice clones had already been downloaded by countless numbers of the company's customers and could still be used.
Mr. Lehrman also questioned whether the company had used the couple's voices along with several others to build the underlying technology that drives its voice cloning system. Speech synthesizers often learn their skills by analyzing thousands of hours of spoken words, in the same way that OpenAI's ChatGPT and other chatbots learn their skills by analyzing large amounts of text collected from the Internet.
Lovo acknowledged that it trained its technology using thousands of hours of recordings of thousands of voices, according to correspondence in the lawsuit.
Mr. Case, the lawyer representing Lovo, said the company trained its artificial intelligence system using audio from a freely available database of English recordings called Openslr.org. He did not respond when asked whether voice recordings of Mr. Lehrman and Ms. Sage had been used to train the technology.
“We hope to regain control over our voices, over who we are, over our careers,” Lehrman said. “We want to represent others who this has happened to and those who it will happen to if nothing changes.”