ChatGPT maker OpenAI has unveiled its latest generative artificial intelligence tool, one that can make audio mimicking real human voices.
OpenAI on Friday shared samples from early tests of an AI tool called Voice Engine, which uses a 15-second sample of someone speaking to generate a convincing replica of their voice. Users can then provide a paragraph of text and the tool will read it in the AI-generated voice.
There are several AI-generated voices services already available to the public but, as it did with the breakout chatbot ChatGPT, OpenAI has proven particularly adept at garnering widespread adoption of AI tools.
An AI-enabled text-to-voice tool could help with translation, reading assistance for children or aiding people who have lost the ability to speak, the company says. But some skeptics worry it could also fuel the creation of disinformation or make it easier to perpetrate scams.
OpenAI says Voice Engine is currently being used by only a “small group of trusted partners,” including education and health technology companies, and it will use their tests to determine whether and how to allow more widespread use. Those testers have agreed not to recreate people’s voices without their explicit consent and to clearly identify to listeners that what they’re hearing is AI-generated, according to the company.
“We recognize that generating speech that resembles people’s voices has serious risks, which are especially top of mind in an election year,” OpenAI said in a blog post. The company acknowledged the need for major changes as AI-generated audio becomes more widely available, although it doesn’t plan to release Voice Engine to the public immediately. For example, the company suggested phasing out voice-based authentication for bank accounts.
“Any broad deployment of synthetic voice technology should be accompanied by voice authentication experiences that verify that the original speaker is knowingly adding their voice to the service and a no-go voice list that detects and prevents the creation of voices that are too similar to prominent figures,” OpenAI said.
Voice Engine can use a voice sample in one language to create a replica voice that can speak in multiple other languages.
Its blog post includes an example of an audio clip of a human reading a passage about friendship, alongside AI-generated audio that sounds like the same person reading the same passage in Spanish, Mandarin, German, French and Japanese. In each of the AI-generated samples, the tone and accent of the original speaker is maintained.
The preview of Voice Engine comes as users await the public release of Sora, the AI-generated video tool that OpenAI teased last month. Sora can create realistic looking 60-second videos from text instructions, with the ability to serve up scenes with multiple characters, specific types of motion and elaborate background details. OpenAI’s ChatGPT can also generate images from a text prompt.