AI cloning voices is worrying. But if it belongs to OpenAI it is even more so

There are already several AIs that allow you to clone a person’s voicebut the fact that he’s working on it OpenAI change everything.

Very soon more than one hundred million users – they are the ones who use ChatGpt every month – they could have a new artificial intelligence model at their disposal that allows you to perfectly imitate any voice after listening to it for just 15 seconds.

The tool is called Voice Engine and OpenAI has been working on it since late 2022. It can use the cloned voice to read a text in the same language it was written in or in a different language. With results, as you can imagine, excellent.

OpenAI has allowed some “trusted partners” to test the new technology since the end of last year. Among these there is Age of Learning, a company dedicated to improving children’s learning. And it used OpenAI’s text-to-voice model to reach a wider audience by translating the original English speech into different languages.

Artificial intelligence

History of ElevenLabs, the startup that clones the voice born from intolerance for film dubbing

by Gabriella Rocco

January 23, 2024

But as is often the case with AI, tools that offer new opportunities can also pose a threat to society. OpenAI itself recognizes this, who writes on his official blog:

“We recognize the serious risks associated with generating synthetic voices that mimic human voices, risks that are particularly relevant in a time of elections [a breve ci saranno quelle per il parlamento europeo, mentre il prossimo novembre sarà la volta delle presidenziali Usa, nda]. To ensure the responsible development of this technology, we are collaborating with U.S. and international partners from government, media, entertainment, education, civil society and more.

Partners currently testing Voice Engine have agreed to our usage policies, which prohibit impersonation of individuals or organizations without consent or legal basis. Additionally, our terms with these partners require the original individual’s explicit and informed consent, and we do not allow developers to create tools that allow end users to generate personal entries.”

The proliferation of synthetic and credible voices it raises the possibility that AI will be used to produce deepfakes, in this case audio of a person appearing to say things they never actually said. As happened some time ago, a Joe Biden.

In fact, last February, several New Hampshire residents received a phone call with a recorded message from the US president inviting them not to go to the polls to participate in the Democratic Party primaries. But the communication turned out to be a deepfake: Biden’s voice was cloned thanks to artificial intelligence.

Following this episode, the US government banned any type of phone calls it involves a pre-recorded message from an AI-generated voice.

Then there is the possibility that an increasingly simple production of synthetic voices – existing or invented – could replace those of human beingsparticularly in the workplace.

The voice actress (and musical star)’s tweet – which went viral – has caused a lot of discussion lately. Sara Poyzer which contained the email sent by a production company to his agency.

It said: “Sorry for the delay but the BBC has given us an AI-generated voice so we won’t need Sara anymore.”

The English broadcaster then released a note on the incident, to try to explain that Poyzer had not been replaced by “just any” voice.

“We are making a very sensitive documentary – reads the BBC statement – which features a participant who is dying and currently unable to speak. We worked closely with his family to find the best way to represent this person’s voice at the end of the film, when a text of him will be read.”

“In these very particular circumstances and taking into account the family’s wishes, we agreed to use artificial intelligence for a short section to recreate a voice that unfortunately can no longer be heard. This will be clearly signaled in the film.”