AI revives dead actor's voice to read audiobook

AI For Business


Getty Images

On the left is Judy Garland and on the right is Burt Reynolds.



CNN

Although actress Judy Garland never recorded her own voice to read the audiobook of The Wizard of Oz, people will soon be able to hear her recite the children's novel that inspired the film.

Earlier this week, AI company ElevenLabs released a digitally-created The company has added narration from famous late actors, including Garland, James Dean and Burt Reynolds, to its newly released Reader app, which the company says will convert any text on your phone, including articles, PDFs, ePubs, newsletters and e-books, into narration.

“We deeply respect their achievements and are honored to have their voices as part of our platform,” said Dustin Blank, head of partnerships at Eleven Labs. “Adding them to our growing list of narrators marks a major step toward our mission of making content accessible in every language and voice.”

The company said it had struck a deal with the actors' estates to acquire the voice actors' copyrights, but gave no details on the compensation. The move shows what artificial intelligence could bring to filmmaking. This sets a precedent not only for Hollywood, but for licensing and working with estates, and at a time when technology has advanced exponentially around the ability to create images, text and audio, making it easy for anyone to voice something someone never actually said.

This raises questions about how artificial intelligence can or should be used in creative industries such as journalism and film.

ElevenLabs made headlines earlier this year when its tools were reportedly used to create fake automated calls from President Joe Biden urging people not to vote in New Hampshire's presidential primary.

The partnership with the stars' estates comes two months after ChatGPT maker OpenAI came under fire for releasing a synthetic voice that bore an uncanny resemblance to Scarlett Johansson's character in the film “Her.” In a statement shared with CNN, Johansson said she was “shocked, outraged, and in disbelief” that the company used her likeness in the synthetic voice. She turned down the opportunity to partner with OpenAI.

While people can't copyright their own voices, they can copyright recordings, said David Gunkel, a professor of communications at Northern Illinois University who tracks AI in media and entertainment: AI is trained on old recordings, and those recordings are copyrighted.

“Eleven Labs” “All new partnerships are within the bounds of the law,” he said. “The estate will make significant amounts of money from licenses and agreements, similar to the companies that negotiate copyright deals to use Queen's popular songs in advertising. Record companies could, in theory, turn down any offer of money.”

Vern Elliott, a vice president and analyst at market research firm Gartner, said AI models can now be trained with fewer voice recordings, and only a few recordings of a celebrity or an average person are needed to capture elements such as tone and style.

“The bigger concern is determining what the owners of those recordings can or can't do to monetize the audio,” he said.

Media companies are also using AI for narration: Last week, NBC announced that an AI version of famous sports anchor Al Michaels will return to the Olympics this summer to report daily on the network's streaming platform Peacock. An NBC spokesperson told CNN that Michaels has been paid to participate.

However, it is unclear how AI versions of famous voices will be received by general audiences and whether they will raise concerns about authenticity.

“We don't know yet what the likely market for this kind of thing is, but we're already seeing that audiobooks are a popular product when they're read by well-known voice actors or celebrities,” Gunkel said. “If there was a way to get celebrities to do all kinds of content without them doing it themselves, that could open up a much larger market.”



Source link

Leave a Reply

Your email address will not be published. Required fields are marked *