Two years ago, technologists at OpenAI published a blog post detailing how the company had developed the ability to replicate human voices using powerful artificial intelligence technology. The new software was so advanced, the researchers said, that OpenAI decided not to release it to the public out of an abundance of caution.
While the company’s position has not changed, it has not stopped OpenAI from continuing to work on the technology. This year, OpenAI quietly purchased a small start-up, Weights.gg, that offered A.I. tools to create clones of people’s voices, two people familiar with the deal said on the condition of anonymity because they were not authorized to speak publicly.
Terms of the deal, which has not been publicly reported, could not be learned. People familiar with the acquisition said OpenAI had bought both the company’s small team of employees and its intellectual property. Weights.gg announced it was shutting down its services in March.
Weights.gg worked as a kind of social network for creating and sharing A.I. algorithms that could do things like clone voices using the company’s free consumer app, which is called Replay.
One YouTube user showed how he had used Weights.gg’s technology to clone the voice of the actor Samuel L. Jackson. Some of the top voice models in the company’s repository included clones of Taylor Swift, Kanye West and members of the popular K-pop band Blackpink.
Weights.gg’s tools were also used to clone other copyrighted voices, including Bugs Bunny and Daffy Duck. Weights.gg’s library also contained the voices of prominent political figures, including President Trump and former President Joseph R. Biden Jr.
Some high-profile figures, including Mr. Jackson, have opposed the cloning of their voices using the technology. Last month, Ms. Swift filed a series of applications with the U.S. Patent and Trademark Office in order to trademark her voice and likeness.
Weights.gg employed a half-dozen people and has raised roughly $4 million in venture capital, according to data compiled by PitchBook, a research firm.
OpenAI has run afoul of copyright issues in the past. Last year, the company released Sora, a smartphone app that allowed people to instantly generate videos of copyrighted characters without their permission. The company quickly ran into resistance from Hollywood before striking deals to use them.
(The New York Times has sued OpenAI and Microsoft, claiming copyright infringement of news content related to A.I. systems. The two companies have denied the suit’s claims.)
It is unclear what OpenAI plans to do with the Weights.gg team and technology. But Weights.gg employees have disbanded to work on teams across different parts of the company, the two people said. It is unlikely to release a similar product to Weights.gg, they said.
In recent months, OpenAI has scaled back some of its ambitions in order to focus on revenue-generating products, as it looks to begin trading as a public company by the end of this year. OpenAI shuttered the Sora video app this year. It has also sought to repair its ties to Hollywood by poaching Instagram’s “celebrity whisperer,” Charles Porch, to smooth over frayed relationships.
Instead, it has focused on incorporating its voice technology into other parts of the company. This month, OpenAI released information on how third-party developers can use the company’s application programming interface, or A.P.I., to incorporate OpenAI’s voice technology into outside apps and services. Those uses, which developers would pay to use, could include providing real-time voice translation services or interacting with “agents” using voice commands.
OpenAI has also released its ChatGPT in Apple’s Car Play app, which allows people to give the chatbot voice commands while driving. That functionality has improved over time through OpenAI’s investments in voice technology.
OpenAI appears to have no immediate plans to release its voice cloning technology outside of a limited set of partners, according to the company’s 2024 blog post. And in other blog posts, OpenAI has said it encourages increased safety practices around voice-cloning technologies like the kind it has created.
