OpenAI's Johansson gaffe puts voice cloning technology in the spotlight

Voice cloning technology was put in the spotlight last week when OpenAI was forced to apologize to actress Scarlett Johansson for using her voice, or a close imitation, in its latest chatbot.

OpenAI denied that the voice it used was Johansson's, but its claims were not supported when CEO Sam Altman slammed the new model on social media in a one-word message using “her.”

Johansson voices an AI character in “Her,” which Altman has previously said is his favorite film about technology.

From the get-go, the AI ​​voice clone proved problematic.

Last year, British company Elevenlabs released voice cloning software that made waves for all the wrong reasons.

Internet pranksters quickly began circulating deepfakes of celebrities, including one showing Harry Potter star Emma Watson reading Hitler's “Mein Kampf.”

Law enforcement agencies have warned that AI clones could be used to extort money from loved ones over the phone.

This technology has developed rapidly over the past year, becoming much more realistic and subtle.

Victor Liparberg, a Danish entrepreneur and CEO of British AI company Synthesia, told AFP that this was mainly due to a program called “Tortoise”, released two years ago.

The program's developers fed thousands of hours of audio data in unstructured format into the model and found that it learned not only what to say, but also how to say it.

“It was a pretty big paradigm shift,” Riparberg said on the sidelines of the VivaTech conference in Paris last week.

Tortoise is an open source program and Elevenlabs was the first company to use it to go to market.

OpenAI uses a similar system, but the details have not been made public.

'It's not very well'

Much of the controversy over voice cloning centers on concerns about people misusing the software.

But the lawsuit against OpenAI is unusual because the company itself is being accused of botched handling of the matter.

“It's very unfortunate that OpenAI did that, it's really not nice,” TALKR.ai CEO Katia Rein told AFP at VivaTech.

“If they really did clone her voice without her knowledge, I think that's very wrong,” Liparberg said.

The two entrepreneurs are among hundreds of others using AI voice programs in ways they claim can make businesses more efficient.

Laing's company offers virtual voice assistants, or AI customer service agents.

He said the company's system can currently resolve 25 to 30 percent of calls without human intervention.

Synthesia specializes in video avatars, which Riparbelli says allow any office worker to turn text and slides into lifelike, AI-acted videos.

Both Riparbelli and Laine allow clients to use their own avatars, pre-made products, or avatars provided by the likes of OpenAI and Elevenlabs.

Lipalberg said the actors' likenesses and voices used by Cynthesia were licensed for two years, with an option to renew after the initial period.

Problems arise when an actor's voice is used without their consent.

“A strange precedent”

The debacle overshadowed a developers conference in Paris last week where OpenAI unveiled a range of new tools.

In front of the auditorium's large screen, Roman Huet, OpenAI's head of developer experience, was chatting briskly on his cellphone.

After a few seconds, his short audio sample was processed and a voice commented in five languages ​​over the resulting video.

The demonstration showed how rapidly the field is advancing, but the headlines have already been written.

“How stupid is OpenAI?” The Washington Post asked in a newsletter, while other commentators suggested that the genius Altman was nothing more than a con man.

Still, Lipalberg accepted OpenAI's claim that it used a different actor with a similar voice to Johansson.

“If it's someone who's not her, but sounds a lot like her … where do you draw the line?” Riperberg asked.

“It would set a very strange precedent if they weren't allowed to use someone who sounds similar to her.”

© 2024 AFP