correct. English only, as @Bayesian pointed out. (clarification added due to ambiguous manifold ui)
I already use AI narration to read books, both fiction and nonfiction. It's at the stage that it sounds natural enough that I forget that it's automated, but also has less expressiveness than a human narrator.
I also listen to the Project Lawful podcast, which is entirely AI generated. Despite being all AI voices, it has surprisingly high emotional expressiveness. It didn't take me long for them to just feel like "people talking" instead of AI's talking. I'm only pulled out when the AI doesn't misparses how to read something in the text (like a "#", which sometimes should be read "number" and sometimes should be read "hash")
Arguably, we're already at the "mostly unnecessary" point. It seems pretty likely that we'll be passed the "narrator Turing test" by 2026.
Alright, I am extremely bullish on the fact that AI will free us from work. However, I listen to a LOT of audiobooks and I can tell you that current voice models would be a major turn-off for me.
A good audiobook is not just someone who reads the words but someone who actually performs. It's why the best audiobooks are done by professional voice actors and not by some random guy or celebrity.
Current audio models, even the new ones displayed by OpenAI, have a certain "falseness" to them. Maybe that's just in my mind, but having listened to so many people read so many books, I feel like I can sense when the intonation just isn't quite right. This will undoubtedly improve in the future, but I am unsure if that will happen before 2026.
The most realistic scenario I can see is that you can have a book narrated in whatever style you like, but that there's some "officialy read" version published as well. If this happens, we'll also surely see this in the indie writing scene on sites like RoyalRoad.
@PickleManiac The dramatic voice acting is exactly why I dislike human narrators. I don't want to hear someone's interpretation of the text, I just want to hear the text. Like a book, but using audio.
@singer Yeah, that's understandable. If you're looking for automatic text narration, we have been here for a long while already. I should mention that I was talking about novels, though. I hardly read non-fiction, haha.
The resolution criteria here is too ambiguous.
I agree with @RomanHauksson comment below. Having books read by the author will likely still demand a premium.
As an example: Barrack Obama's next memoir will probably come out within a year or two. Obama is one of the most deepfake-able people alive, with peak AI quality available. Yet, I suspect consumers will have a strong preference for him actually reading it. For him and other memoir authors, how they choose to intonate the moments and people of their life all make up their personal story. These are things AI and listeners can guess, but without the author's telling cannot know.
But again, the resolution criteria is too ambiguous. One could say that "mostly unnecessary" means we can carve out and exclude memoirs or a few other categories. One could also say "mostly unnecessary" means AI just needs to get greater than 50% of the quality or understandability. But if these exceptions are allowed for "mostly unnecessary", you could probably resolve this "yes" now.
If defined in terms like "will a majority of the 100 best selling audiobooks in 2025 be read by AI", I think "no" is a pretty safe bet.
Until a better definition is given, I'll hold my Mana.
Listen to this Audiobook by Tyler Cowen https://www.youtube.com/watch?v=dPr4DL4h7Wg (released yesterday)
It's AI, he said his wife and daughter couldn't tell the difference. I definitely couldn't
@GazDownright I was betting on this: "if AI narration becomes indistinguishable from human narration to the untrained ear, it would resolve as YES.". I agree it should be placed in the description as well tho.
@patrik Okay, cheers. I have a perhaps nitpicky but not unthinkable follow-up in light of the Sam 'Not Consistently Candid' Altmann and Scarlett Johansson situation. What if legislation prohibits AIs from having specific voices from individuals who have not given away their rights? You have the same with image rights today. This means there will be a market where living individual human voices are still sought after, thus not redundant. How would you resolve this admittedly hypothetical case?
@GazDownright I am not the creator of this market. But I imagine that it doesn't have to mimic a specific person's voice.
@FranklinBaldo can you be more specific as to what "mostly unnecessary" means? I think many people in social spheres that tend to hype AI will be claiming it is mostly unnecessary—indeed, some would already claim that—but I don't the audiobook narration industry will be reduced by more than, say, 80% (measured by dollars, people, or jobs). Just as with other industries (e.g., radiologists, graphic design), AI hypers tend to underestimate the challenges, such as the large practical difference between 99% and 99.9% "performance" and the hard-to-notice subtleties involved in human skills (e.g., Polanyi's paradox).
@Jacy even if human narration continues to exist, if AI narration becomes indistinguishable from human narration to the untrained ear, it would resolve as YES.
@FranklinBaldo that doesn't seem consistent with the question posed, but sure, I would guess that an untrained population (e.g., US nationally representative crowdworkers) will still be able to distinguish the average AI narration from the average (professional) narration with statistical significance (e.g., 100 workers each with 10 tests each of 1 minute of complex book-style audio).