OpenAI’s Human Voice Recreation Tech Unveiled—But You Can’t Have It

OpenAI is sharing its Voice Engine technology with testers and trying to gauge its dangers. The tech can recreate someone’s voice with just a 15-second recording of them reading a paragraph to create a synthetic voice that sounds like the reader’s. The voice doesn’t have to stay within the confines of the reader’s native language.

OpenAI is trying to understand its potential dangers, such as spreading disinformation across social media, impersonating people online or during phone calls, or breaking voice authenticators’ ability to control access to online banking accounts and other applications. The company is looking at watermarking the synthetic voices or adding controls that prevent people from using the technology with the voices of politicians or other prominent figures.

OpenAI does not offer a public tool allowing individuals and businesses to recreate voices from a short clip like Voice Engine. They admit that the technology could be hazardous in an election year.

On the positive side, the tool could be helpful to people who lost their voices through illness or accident, and businesses can use the technology for audiobooks, online chatbots, and an automated radio station DJ.

OpenAI is not alone in generating synthetic voices; Google and ElevenLabs are also looking into this tech. OpenAI says it has yet to make any plans to make money from the technology.