Voice Engine: OpenAI unveils new tech tested to recreate human voices
3 Min Read
"Voice Engine" is not being shared more openly at the moment because the company is still trying to test its potential dangers.
-
The OpenAI logo is displayed on a cell phone with an image on a computer screen generated by ChatGPT's Dall-E text-to-image model, Dec. 8, 2023 (AP)
Starting with an AI tool that enables people to create digital images simply by describing what they wish to see and then moving on to a technology that generates full-motion video, OpenAI has now offered a new tool capable of recreating a human being's voice
Voice Engine, the company said, is being tested by a small group of businesses to recreate someone's voice from a 15-second recording.
The voice doesn't have to be in the native language as it can recreate the voice from English for example into Spanish or French.
It added that it is not being shared more openly at the moment because it is still trying to test its potential dangers like spreading disinformation across social media, allowing criminals to impersonate people online or on calls, or, in particular, being used to breach voice authenticators for access to online banking accounts and other personal applications.
That being said, it's seeking ways to watermark synthetic voices or include controls to ban people from using the technology with the voices of politicians or other figures.
Read more: Google pauses generative AI 'Gemini' over image inaccuracies
During an interview, OpenAI Product Manager, Jeff Harris, said, “This is a sensitive thing, and it is important to get it right.”
A 'voice' for the 'people'?
Other tech giants like Google and start-ups like the New York-based ElevenLabs have created similar voice-generating technologies that could generate audiobooks, give voice to online chatbots, or establish an automated radio station DJ.
Harris continued that being able to recreate a voice in any way is what makes the technology dangerous - especially in an election year.
For example, back in January, residents of New Hampshire received robocall messages, which the Federal Communications Commission later prohibited, after it told people not to vote in the state primary but in a voice that was probably artificially generated to sound like US President Joe Biden.
According to Harris, there are no plans or intentions by OpenAI to make money from it, since it could be used to aid people who lost their voices through illness or accident.
He provided one example in which a woman’s voice was lost after brain cancer damage, but now, thanks to the tool, she could speak after providing a recording of a presentation she made once in high school.
Last month, OpenAI released a program that can make incredibly realistic video clips from only a few lines of text, prompting content creators to question if they are the next professions to be replaced by robots.
The technology, known as Sora, has elicited reactions ranging from ecstatic excitement to concern about the industry's future path. However, the technology has not been released to the public yet.
YouTuber Marques Brownlee described it as "frightening" and "threatening" to witness an AI doing his job.
