Hacker News new | past | comments | ask | show | jobs | submit login
Apple unveils suite of AI-voiced audiobooks (theguardian.com)
29 points by tkgally on Jan 4, 2023 | hide | past | favorite | 13 comments



I just listened to a couple of free previews. I don't know how many voices they've got but the female one you could definitely tell was AI, it had the weird stilted voice where the timing is off like Siri. The male voice I'm listening to at the minute is a lot better and I'm not sure you'd be able to tell if you didn't know. Occasionally you get a little bit of that 'robotic digital' sound but it's pretty good. I'm impressed at how it can understand the text and take an appropriate emotional tone such as when a character is experiencing doubt and asking a question.

Female Voice: https://books.apple.com/gb/audiobook/shelter-from-the-storm/...

Male Voice: https://books.apple.com/gb/audiobook/pale-moon-rising/id1640...


I listened to a few samples, too, and my impression is the same: a few sound AI-ish in spots, but often I would not be able to tell that the reader wasn't human. I’ve heard better human-read audiobooks, but I’ve also heard worse.

Most of the books identified as “Apple Books audiobook narrated by a digital voice based on a human narrator” seem to be romance fiction. From what I’ve read about the genre, romance novels are written in huge quantities based on standard story, character, and situation templates, and enthusiastic fans often read dozens or hundreds of the books a year. Romance-fiction audiobooks would therefore seem to be a prime target for good-enough AI production. Presumably the stories themselves will be next.


If I didn't know better, I would not have guessed that the male voice was computer generated. The inflection sounds just like someone reading the book.

I used to use the Kindle's text to speech feature before they removed the speaker, while driving or doing dishes, and that was just the standard robotic voice. This is fantastic, though I agree with another comment that buying it for the normal price is a bit odd.


The voices sound pretty good, better than I expected. I “recognize” the female voice. It sounds a lot like somebody who I’ve heard read many New Yorker or Atlantic articles ether for Audm or the old Audible New Yorker service. I’ll try and find the readers name, but I’d bet anything she was used in the training.


Woah, this female voice is clearly sampled from one of the Audm narrators.


I had the same impression, do you know the name?


This would be fantastic as a free bonus with purchase of the ebook, but I'd be extremely reluctant to actually purchase an AI audiobook. Doesn't seem worth it.


It's certainly better than nothing. But even narrators who don't do character voices bring their own personality to it, which is part of the joy of audiobooks.

And good modern audiobooks are a full-on performance. I love Patrick Rothfuss' books on their own, but Nick Podehl brings them to life in a whole new way. You can't AI that, not with anything resembling current technology anyway.


Will the AI voices be able to...uh...do voices? Good audiobook narrators differentiate the voices of characters to give each one their own tone and personality. Character speech shouldn't use the same voice as the rest of the narration.


> Will the AI voices be able to...uh...do voices?

Anything you can imagine, public figures included. (Example: https://fakeyou.com/)


Audiobook narrators generally find that it's better not to use very different voices, just variants of their own. You can use different speakers to voice different characters in the book, but it's usually distracting and jarring rather than immersive.

Perhaps AI voices (and their directors) will find a new approach and create a whole new kind of narration. Perhaps it would be possible to merge a wider diversity of voices if they're all controlled by a single narrative "thought". That would be nifty.


Finally, I've been waiting quite a while for something like this. Now if only I could use an open source local version myself.


The voices are better than expected. Something like this would be great for novels that would not otherwise get an audiobook.

But its nowhere close to the great pros - Robert Keating, Scott Brick, George Guidall don't just read but give the book and characters a personality. Our models are probably ten years away from that.




Consider applying for YC's W25 batch! Applications are open till Nov 12.

Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: