Audiobooks are a booming business, but it’s impossible to imagine highly skilled artists reading aloud every book ever written. To fill that gap, Audible, Amazon’s audiobook division, is testing the use of AI-generated voice clones to speed up the process. The idea of a massive proliferation of audiobooks is clearly appealing, but it also raises questions about whether they can match human performance—or whether it’s worth pursuing even if they can.
The program, in beta in the U.S., is designed for a group of current Audible narrators to create voice clones using AI trained on samples of their speech. They can then plug these AI replicas into their book recordings. The AI voice won’t just be a plug-in for certain chapters and passages, either. Narrators will be able to edit their AI voices to correct pronunciation, slow down or speed up the pace, and then review the final production to ensure accuracy, just as they would a live reading.
Audible has suggested that AI voices will exist alongside human performances in a book, supporting the narrator rather than replacing them. That means narrators will retain control over their AI voices and will be able to audition their real and synthetic voices for reading projects. The exact payment system Audible will use for the AI voices hasn’t been fully fleshed out yet, but the beta will offer a share of royalties on title revenue. How much narrators can expect to earn from AI-generated recordings compared to live readings is less clear. Additionally, Audible noted that beta participants will be able to produce their AI voices for free, suggesting that those not in the beta may have to pay a fee or possibly a subscription for the synthetic version of their voice.
“Audible continues to hear from members of our creator community who are interested in using emerging technologies to make more audio content available to listeners. There is a vast catalog of books that don’t yet exist in audio, and as we explore ways to bring more books to life on Audible, we are committed to carefully balancing the interests of authors, narrators, publishers, and listeners,” Audible explained in the announcement. “Narrators will still retain control over the projects they choose to audition for, whether through voice replica or live performance, and will remain central to the production process.”
AI bookworms
Assuming the beta test goes well, the option to add an AI-generated audio track will be expanded to authors, agents, publishers and other copyright holders later this year. That suggests Audible is testing both the technical and political waters to see how the people involved in creating an audiobook feel about the idea before it’s available to listeners.
Still, even if you can’t get an AI version of a current professional narrator, AI-generated voices are spreading thanks to improved synthetic voice cloning. For example, ElevenLabs now offers AI versions of the voices of James Dean, Burt Reynolds, Judy Garland, and Sir Laurence Olivier for its Reader App, which can read out everything from books to websites.
Amazon has been clear that it’s focused on making sure the people involved in the project are happy. But the company has been a little more casual with its self-published audiobooks, adding AI voices through Kindle Direct Publishing (KDP). Authors have been able to turn their books into audiobooks read by fully synthetic voices, with mixed reviews for their quality. There are now tens of thousands of titles on Audible with AI-generated voices. Professional narrators have complained about how it degrades their work, sacrificing quality for cheapness while hurting their chances. The lack of a filter for books read by AI hasn’t helped matters. If Audible wants to navigate these issues and appease the writers and narrators, along with the fans who want to listen to the books, the company will have to listen carefully.
