Judy Garland, James Dean, Burt Reynolds — these names all share two things in common.
They're both Hollywood legends and also very dead.
But now, a new artificial intelligence (AI) app is attempting to bring them back to life to... read some emails.
AI start-up ElevenLabs announced this week they would use the voices of legendary Hollywood stars as part of its text reader app.
The voices would be narrating articles, PDFS, ePublications, newsletters or any other text format.
Here's what you need to know.
Who's involved?
So far, ElevenLabs has procured the voices of Judy Garland, James Dean, Burt Reynolds and Sir Laurence Olivier.
It says it partnered with the dead celebrity estates — for an undisclosed sum.
Liza Minnelli, Garland's daughter, says the partnership "will bring new fans to Mama, and be exciting to those who already cherish the unparalleled legacy that Mama gave".
"It's exciting to see our mother's voice available to the countless millions of people who love her," she said in a statement.
Celebrity agency CMG worldwide helped pave the way for the partnership and said the app provided "new opportunities" for its clients.
"We appreciate their thoughtful approach in working with talent in the right way," chief marketing officer Tina Xavie said.
ElevenLabs says the partnership will "deeply respect" the legacy of the celebrities.
The voices will not be incorporated into the company's larger audio database.
Is this even legal?
Hollywood union SAG-AFTRA has recently sponsored a bill that would stop dead performers being replicated in audiovisual works and sound recordings without the consent of their estate.
Los Angeles SAG-AFTRA president Jodi Long said the bill now before the Californian Senate would stop the dead becoming someone's "unpaid digital puppet".
In this particular case, chief scientist at UNSW's AI institute, professor Toby Walsh, says the permission of the estates does a lot to assuage this fear.
Still, there is the copyright to think of.
"These systems are first trained on tens of thousands of hours of audio, so they can speak," Professor Walsh says.
"And then they're fine-tuned to speak in a way that sounds like a particular person (here, a dead actor).
"And often that first step, training on tens of thousands of hours of audio, uses content that is scraped from the web without consent, or compensation and that is likely under copyright.
"So not all of the copyright concerns go away."
Beyond the legal issues, distinguished professor of digital media at Queensland University of Technology (QUT) Jean Burgess is unconvinced the voices could remain true to their purpose.
"It seems unlikely that they would be able to keep control of what happens to the audio that's produced by their model," she says.
"So they may keep the training data locked up for a particular purpose but then if you take the audio book that's produced using that training data, it could be used as training data for other kinds of purposes down the line."
Who else could be affected?
For voice actors, the technology could seriously disrupt their way of life.
Professor Burgess, who is also the director at QUT's GenAI Lab, says the deal makes for "interesting downstream implications".
"Who wouldn't prefer honestly to have Judy Garland read The Wizard of Oz, if it sounds authentically like Judy Garland, than to have an unnamed, perfectly competent professional voice actor read the Wizard of Oz?" she asks.
"(It's) likely to become a harsh reality for many really skilled voice actors who are making up the bulk of their living through this sort of work, and subsidising their other, more artistic and creative activity in that way."
Professor Burgess says the other impacts would depend on audience reaction, which she's "curious" about.
"There's certainly a compelling star quality to those voices," she says.
"I'm curious myself ... but whether or not it turns out to be a short-term curiosity or a longer-term change to the way the whole audio books (industry) works remains to be seen."
Professor Walsh has a more confident view as to fans' reactions.
"Having Taylor Swift read you a bedtime story. Even say your name," he says.
"(It's) sure to be popular."
Who is ElevenLabs?
ElevenLabs says it is an AI audio developer making "realistic, versatile and contextually-aware speech, voices, and sound effects across 29 languages".
You may have also heard of the start-up via its implication in a fake US electoral robo-call urging voters not to vote for Joe Biden.
About 20,000 New Hampshire voters received a call purporting to be from the president, urging them not to vote for him in the January 23 primary.
Voice-fraud detection company Pindrop Security Inc later analysed the call and traced it back to ElevenLabs.
The creator was suspended by ElevenLabs, according to a person familiar with the matter, who spoke to Bloomberg.
Vice also reported last year ElevenLabs had been used to make deepfakes of celebrities like Emma Watson saying racist and transphobic slurs which were spread via 4chan.
How have other celebs reacted to AI impersonation?
To put it lightly, not well.
Just last week, actor Morgan Freeman came out labelling AI generators using his voice "unauthorized".
He also thanked fans for spotting the AI, saying they helped ensure "authenticity and integrity remain paramount".
In May, Scarlett Johansson said she was "angered" to hear an OpenAI chatbot voice that sounded "eerily similar" to her own.
OpenAI later said it had halted the use of the ChatGPT voice.
And 200 artists, including Billie Eilish and Katy Perry, warned in April of a "race to the bottom" if AI remained unchecked.
In Australia, voice actors too have raised concerns about losing work due to AI imitation.
Western Australian voice artist Jordan Fritz told AM he had seen a 90 per cent drop in smaller jobs in the past 12 months as a result of voice cloning.
