Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

The interesting and fascinating thing about audiobooks is that we will likely within the next 1-2 years have essentially complete generative AI produced versions of everything imaginable, meaning that the definition of book and audiobook will become one and the same.


I don't think thats the case.

_good_ audiobooks require the reader to have consistent voices for each character, express the correct emotion for the particular scene, and intone some sort of personality into the book.

A good example is Kobna Holdbrook-Smith (https://www.audible.co.uk/pd/Rivers-of-London-Audiobook/B004... listen to the free sample) He makes the characters different, and the main character PC grant comes alive in his voice.

A negative example is https://www.audible.co.uk/pd/Breakfast-of-Champions-Audioboo... where John Malkovich lethargically stumbles his way through, like a bored supply teacher with a cracking hangover and no lesson plan.

Now, you can make very passable recordings of a book using something like amazon's polly. You can make it more bearable when you inject verbal clues using the markup tool. But. It still feels rubbish listening to it.

However its a shit tonne cheaper, so its probably going to seep in because people want profit more than art, but it would have to be made cheaper for it to take off. There is librevox who've been around for years. They have volunteers to read books and they distribute them with permissive licenses. However thats not as popular as audible. Partly because of content, partly performance.


It's surprising that "cognitive" aspects of human speech seem to be "ahead" of more mechanical aspects like text-2-audio.

Computers are or will be better at writing the book than most humans. They'll still need to hire a human narrator to make it sound good.


I think the 1-2 years estimate is a bit short. Right now, a computer can create a droning audiobook out of a written book. Within the next few years, generative models will create cheap mediocre audiobooks.

But the technology is changing very rapidly. I would expect that by 2030 generative models will have the ability to produce some very good audio. Possibly on par with professional human voice actors.


With AI?

Every character in the book can be voiced in a unique manner, with the narrator matching only the character as needed.

Additionally? Those voices can be tuned for the listener; an AI that has been trained for your preferences well beyond the simple matching that AirPods do today with an audiograph.

If you listen to Audiobooks, try listening to one that has been performed by a cast.


It's not for everyone. Personally I detest "graphic audio" audiobooks, generally because they leave out all the descriptive prose from the book, both denying me the author's writing style and forcing me to trust the way they interpreted his descriptions are the same way I would have. I want to read a book with my ears, not watch a stage play without my eyes.


You are right. Graphic Audio (graphicaudio.net) made me re-listen to Brandon Sanderson's Stormlight Archive (all 4 books). With a cast, it becomes so much entertaining.


I dislike when it’s a full cast, I actually prefer a single person to voice it. Having different voices makes it feel like I’m listening to a movie rather than reading a book.


I wonder. won't it take longer?

People pay for decent quality. I hate listening to artificial voices. I loved audiobooks narrated by wil wheaton.

I remember when kindle books first came out and everybody threw their books through OCR and books were full of typos and extra punctuation.

I suspect audiobooks that are read by synthesized voices will go through the same "uncanny-valley" dissatisfaction.




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

Search: