Cheap AI voice clones may wipe out jobs of 5,000 Australian actors
A voice actor at work. Companies may come to regret the lack of human connection if the voice reading an audiobook is AI-generated, Simon Kennedy says. Photograph: visualspace/Getty Images
Voice actors say they’re on the precipice of their work being replaced completely by artificial intelligence, with corporate and radio roles already beginning to be replaced by cheap generative AI clones.
While a high-profile actor like Scarlett Johansson can make the most prominent AI company in the world back down within a day from using her voice likeness in their AI products, everyday actors working on commercials, audiobook and video games worry they risk having their own voices cloned, or miss out on work entirely due to the rise of AI voice clones.
The Australian Association of Voice Actors (AAVA) told a parliamentary committee investigating AI the jobs of an estimated 5,000 local voice actors are already in danger, with the group pointing to one national radio network actively investing in technology to replace human voice actors.
In its submission, the group criticised the development as “a disappointing move from a player in an industry that has relied on voice artists to bring quality, credibility and humanity to their medium for over 100 years.”
The recently formed association’s president, Simon Kennedy, told Guardian Australia, the advent of AI and its impact on the voice industry was partly the catalyst for setting up the group, but he says they’re “not anti-tech and we’re certainly not anti-AI”. The group, he said, just wants fair rules around how the technology will be used, and protection for people’s voices against being misused by AI.
He said the canary in the coalmine for voice actors will be audiobooks.
“Audiobooks is frontline because of the volume of material and the perceived cost-saving that the companies that create them think that they’ll make.”
He said companies may come to regret the lack of human connection if the voice reading a book is AI.
“When it’s an AI voice, I think they’re going to find people just don’t bother with their audiobooks any more. They are just like, ‘I’m feeling nothing’.”
Kennedy said corporate work and education material were also low-hanging fruit for organisations looking to cut human voice work, but advertising will take longer.
“Big advertisers want quality and AI is not going to be given to them for quite some time.”
He said voice actors selling their likeness had not thought through the long-term impacts.
“I don’t think that the endgame was really top of mind for people; that your voice will now exist in the marketplace, as a digital clone of yourself, that will basically take work that you could normally get yourself,” he said.
Last year, Guardian Australia reported Australian software developer Replica Studios had licensed 120 voices from actors for video game development that will pay a fee to actors when clones of their voices are used in video games. In January the company signed an agreement with the Screen Actors Guild in what it says is an ethical approach to AI voice use – where all content is licensed.
‘Wide as an ocean, shallow as a puddle’
The reaction from actors has been mixed. Cooper Mortlock, an Australian actor who began working in voice acting at the start of the Covid-19 pandemic, said it would undercut work by up and coming voice actors trying to get a foothold in the industry – particularly if they use AI-generated scratch voices as a place holder for the final voice, during the production process.
“It’s not only things like that, and not only limiting opportunities for the artists themselves, but also the creative scope of the projects,” he said. “There’s no opportunity for happy accidents or surprises – because AI is taking already existing things and just repurposing [them].”
He said using AI voices to generate dialogue will lack the creativity that comes with using a human voice actor on scripted dialogue.
“It’s as wide as an ocean but as shallow as a puddle,” he said.
“You compare something like some recent video games that are very focused on story narrative and character like Cyberpunk 2077, The Witcher 3, Baldur’s Gate 3 … those games are so meticulously crafted.”
Up until now, AI voice clones often struggled with non-American accents. Australian voices, for example, often keep an American inflection. Newer services now offer a full suite of different accents of Australians at different ages. Kennedy said he hoped the delay was a sign Australians were holding out giving over their voices.
“We’re holding back until there are ethical frameworks in place where we can license our voice knowing that we are going to be treated fairly and compensated fairly,” he said.
But Mortlock said the lag was due to Australia being a smaller market, without the big dataset for the AI to learn the nuances of Australian voices.
“There’s more data available now. I think it was [a] very US-centric thing … It’s just expanding – I don’t necessarily think it has to do with the accent itself.”
The AAVA has called for laws to govern consent, control and compensation around how AI voices are used, making sure artists are paid fairly and have complete control how – if licensed – their voices are used.
Mortlock has said an animation project he worked on cloned his voice and used it without consent after the work stopped – something the company he worked for denies – and part of the issue is there is no transparency when AI is being used.
He would like AI banned from the creative industry to ensure workers could remain employed, but said a tax on the use of technology to compensate workers, as well as greater transparency, would be appropriate.
“The actors should be reimbursed and I think there needs to be disclosure as to who each voice is they’re hiring. Because otherwise they could take this actor off the internet … it’s become the ‘wild west’.”