麻豆官网首页入口

People can choose to sell their voice

Towards the end of his life, James Earl Jones made the decision to sell the rights to his voice, meaning his vocal likeness can be used by companies that hold or have licensed those rights. Software can learn his voice and recreate it.

It’s not just actors being brought back by AI. Benjamin Field, of AI company Deep Fusion, is the producer of Virtually Parkinson, a podcast that resurrects the late legendary chat show host Sir Michael Parkinson. “We needed an authentic representation of Sir Michael’s voice in order to perform unscripted interviews with new guests,” says Field. “We licensed approximately 100 hours of Sir Michael’s voice from Parkinson Productions and fed those into a synthetic voice model, which then learned what Sir Michael’s voice sounded like.”

The potential for banking famous voices is huge, but the key here, says Field, is that it’s done “legally and ethically”, always with the permission of the person’s estate, “otherwise we destroy our own industry and the trust we have with our audience.”

Could AI voice acting create more jobs, not take them?

Let’s clear up the two different ways AI voice synthesising works: “One is a text prompt,” says Benjamin Field. “So you can build a synthetic voice model that will autonomously decide how the intonation of that voice flows across a piece of dialogue.” In short, you write dialogue and the AI software decides how to read it. That’s the cheaper, less convincing way of doing it. “Or you can have a voice-to-voice model. That requires a human voice that will generate the intonation and expression. Then James Earl Jones’s voice will be dubbed on top of that.” So, an actor gives a performance, then their voice is made to sound like someone else’s.

One of the major criticisms of AI is that it means computers take jobs from people. Field’s argument is that voice-to-voice creates more jobs, because it not only requires an actor, but also sound engineers. And he says AI can help living actors make more money. “It almost makes you more available for work, because you don’t always have to be present for it.”

Is an AI performance actually a performance?

“I heard Tom Hanks say in an interview once… that after he’d passed he would be able to go on performing,” says Field. He believes that AI allows beloved actors to be part of new projects after death.

Impressionist Alistair McGowan

But Aleks Krotoski asks if we can actually consider it the actor’s work if they’re not making any of the acting choices. “Isn’t it about the performance?,” she says. She asks, “Does using this technology kind of fossilise the character? Does it allow that character to evolve, or does it mean it can never change?”

If Tom Hanks’ voice is used in a movie after his death, is that still a Tom Hanks performance?

Sounding like someone is harder than it seems

AI is not the only way to recreate the voice of a dead performer. You can, of course, hire another actor to imitate them. Famously, the voice of Wallace in Wallace and Gromit is now performed by Ben Whitehead, following the death of the original Wallace, Peter Sallis, and you’d be hard pushed to tell the difference.

Alistair McGowan is one of the UK’s best-known impressionists. He’s sometimes been hired to imitate dead people for voice-over purposes. He narrated the documentary My Name Is Alfred Hitchcock as Alfred Hitchcock. He says recreating a person is not just about the sound of the voice, but the behaviour, which AI may be less able to capture. “It’s about the person,” he says. “It’s about the character… With Hitchcock, it’s about the noises he makes in between things… You have to know the character, not just the voice.”

McGowan mentions that one of his favourite voices to do is his dad’s. “My father died in 2003 and I met my wife in 2011, so she never met my father,” he says. “But because I can do my father’s voice… she says, ‘Now I know your dad.’” And nobody could do that but him.

AI doesn’t have soul

Alistair McGowan says the human brain is trained to detect minute changes in people’s voices. “Even if somebody is slightly unwell or slightly down… you’d pick it up within five seconds and say, ‘Are you all right? You sound different’… Everything is reflected in the voice. Now, can AI replicate that human soul?”

He says that convincingly recreating a voice is “like classical music… it’s about the gaps between the notes… When I watch good acting, it’s the same thing. It’s about the gaps between the words.” And that requires feel, not just computer logic.

So perhaps AI can save Darth Vader, says Kevin Fong, because Darth Vader has a specific, emotionless sound. He’s already a synthesised voice. “He’s easier to imitate, perhaps easier to preserve.” But can AI save any other James Earl Jones performance? That’s a trickier question.