Did anyone else get a notice that they're eligible for this? From those who have, any first impressions? I am not currently wowed.
-
Welcome to the Fantasy Writing Forums. Register Now to join us!
Amazon audiobooks AI beta
- Thread starter skip.knox
- Start date
pmmg
Myth Weaver
I have not, but I have not looked at my amazon account for a while.
Generally, I think any effort at this would most likely be close enough, but less than perfect, but...for me, its a non-starter. I dont just want it read aloud, I want the voice readers to get it right.
Generally, I think any effort at this would most likely be close enough, but less than perfect, but...for me, its a non-starter. I dont just want it read aloud, I want the voice readers to get it right.
Dylan
Troubadour
Don't think so?
I don't even have an amazon account
I don't even have an amazon account
Prince of Spires
Istar
I had a look around, but I don't have it (yet). Then again, living in the Netherlands, I also don't have access (as an author) to Audible, so there's that small issue which might explain it...
I did! It didn't take long either to put the rest of my titles up on Audiobooks (I already had a few on there). I would for sure do them so you have other ways to make a sale.
- Thread starter
- #6
I've listened to whole chapters from my novels and I just cannot pull the trigger. Here are some of the problems I encountered.
1. Stuttering. A character is freezing cold and says "p- p- please", which the AI renders as Pee Pee Please.
Not ideal.
2. Fun with words. In another novel a pompous man is named Appius Aetius Arvinus, which my MC at once puckishly renders as Aaa (to be pronounced "aaah" of varying lengths). The AI renders this as "Triple A". The AI is, rather surprisingly, a fan of American baseball.
3. Inconsistent pronunciation of invented words. The dwarvish captain of a construction camp has the title "Vorgeer" (German readers will recognize the source here). Sometimes the AI gives this as "Forjeer" and sometimes it's "Forger" (hard g). One of my main characters is Tusco, who is sometimes Toosco and sometimes is Tusk-o. And this doesn't even enter into frequent use of other actual languages in my books (Latin, French, German, Italian).
There are many other examples. They are obvious goofs and I just cannot have my readers giggling every few pages. Detracts from the mood.
Now, Amazon support--which has been prompt in its replies (probably also AI)--tells me I can fix these problems by editing the pronunciation, which is possibly true, but the editing function has so far proved utterly unreliable. Sometimes I can edit and sometimes the option is grayed out. Mostly it's been the latter. And anyway the change in a name, for example, does not appear to pass all the way through the book. I would have to change every occurrence. Also not ideal.
But let's imagine all that is fixable. I'm still left with odd emphasis placement and changes in pitch. I can't be sure yet, but it sounds like the narrator pitches their voice up for a female speaker, but only when there's an obvious indicator such as "she said". Other times, the pitch is the same as for a male speaker. That inconsistency is the distraction.
All in all, what AI can achieve in narration is truly remarkable. It's far more sophisticated that what I would have guessed possible even a year ago. But I'm not ready to have my stories rendered (a fitting verb) by AI quite yet. And this leaves aside (for now) issues with pricing, terms and conditions, and other publish-side factors.
I can see the future, but I don't like the sound of it.
1. Stuttering. A character is freezing cold and says "p- p- please", which the AI renders as Pee Pee Please.
Not ideal.
2. Fun with words. In another novel a pompous man is named Appius Aetius Arvinus, which my MC at once puckishly renders as Aaa (to be pronounced "aaah" of varying lengths). The AI renders this as "Triple A". The AI is, rather surprisingly, a fan of American baseball.
3. Inconsistent pronunciation of invented words. The dwarvish captain of a construction camp has the title "Vorgeer" (German readers will recognize the source here). Sometimes the AI gives this as "Forjeer" and sometimes it's "Forger" (hard g). One of my main characters is Tusco, who is sometimes Toosco and sometimes is Tusk-o. And this doesn't even enter into frequent use of other actual languages in my books (Latin, French, German, Italian).
There are many other examples. They are obvious goofs and I just cannot have my readers giggling every few pages. Detracts from the mood.
Now, Amazon support--which has been prompt in its replies (probably also AI)--tells me I can fix these problems by editing the pronunciation, which is possibly true, but the editing function has so far proved utterly unreliable. Sometimes I can edit and sometimes the option is grayed out. Mostly it's been the latter. And anyway the change in a name, for example, does not appear to pass all the way through the book. I would have to change every occurrence. Also not ideal.
But let's imagine all that is fixable. I'm still left with odd emphasis placement and changes in pitch. I can't be sure yet, but it sounds like the narrator pitches their voice up for a female speaker, but only when there's an obvious indicator such as "she said". Other times, the pitch is the same as for a male speaker. That inconsistency is the distraction.
All in all, what AI can achieve in narration is truly remarkable. It's far more sophisticated that what I would have guessed possible even a year ago. But I'm not ready to have my stories rendered (a fitting verb) by AI quite yet. And this leaves aside (for now) issues with pricing, terms and conditions, and other publish-side factors.
I can see the future, but I don't like the sound of it.
- Thread starter
- #7
But I'm still very much interested to hear from anyone else in the beta program and their own experiences and perspectives.
Prince of Spires
Istar
I think it's the curse of the Fantasy genre to be hard for AI to read. There are a lot more invented or unusual words in a medieval-esque Fantasy tale than in a modern day Romance novel set in New York or some undefined American town.
I'm guessing that once the editing options become better, it will more turn into a lot of work that's cheaper than hiring a human narrator, but not necessarily faster.
As for how it sounds, I guess part of that is reader preference. I know some people prefer to listen at double speed (I sometimes do this with Youtube videos if I just want some information quick), which removes a lot of the nuances and decreases the difference between human and AI voices (though it doesn't completely remove them). I still prefer human voices, but the difference is getting smaller, and the price difference is pretty huge, at least compared to the number of sales I would be expecting. At my current level of sales, I would be looking at 500+ years of sales to break even on a human narrator...
I'm guessing that once the editing options become better, it will more turn into a lot of work that's cheaper than hiring a human narrator, but not necessarily faster.
As for how it sounds, I guess part of that is reader preference. I know some people prefer to listen at double speed (I sometimes do this with Youtube videos if I just want some information quick), which removes a lot of the nuances and decreases the difference between human and AI voices (though it doesn't completely remove them). I still prefer human voices, but the difference is getting smaller, and the price difference is pretty huge, at least compared to the number of sales I would be expecting. At my current level of sales, I would be looking at 500+ years of sales to break even on a human narrator...
- Thread starter
- #9
The invented words haven't been the biggest challenge. More difficult has been getting the emphasis right. If I have a character say something like "Oh, I didn't think you meant me.", the AI reads this in whatever tone it was employing at the time. It's been fairly good at getting the tone right for a question, and even sometimes light-hearted or serious. But a sentence like this one can be read in multiple ways. I could emphasis "think" or "meant" or "me" and the connotations would be different. Not for the AI, though.
The editing options definitely have room for improvement. I'd love to be able to use different voices for different characters; right now, I can only choose voices at the chapter level. I'd love to be able to pick from a range of tones (ominous, joking, dramatic, frightened) and apply them at the phrase level. I do think that will come with time.
Meanwhile, it's a very close choice for me. I was able to audition voices and make a choice within an hour or so. Working with humans, that would have taken weeks if not longer. With this beta, I still have to listen to the whole thing, making adjustments now and again, but at the end of it, if I wish, it gets published. Again, far faster than working with humans.
There are still shortcomings. Some are serious enough to make me hesitate. But for me, it's pretty much either this or forget about having my books in audiobook format. I know there are people who cannot read (dyslexia or other physiological problems). There are still more who will not. So it's a tough choice.
The editing options definitely have room for improvement. I'd love to be able to use different voices for different characters; right now, I can only choose voices at the chapter level. I'd love to be able to pick from a range of tones (ominous, joking, dramatic, frightened) and apply them at the phrase level. I do think that will come with time.
Meanwhile, it's a very close choice for me. I was able to audition voices and make a choice within an hour or so. Working with humans, that would have taken weeks if not longer. With this beta, I still have to listen to the whole thing, making adjustments now and again, but at the end of it, if I wish, it gets published. Again, far faster than working with humans.
There are still shortcomings. Some are serious enough to make me hesitate. But for me, it's pretty much either this or forget about having my books in audiobook format. I know there are people who cannot read (dyslexia or other physiological problems). There are still more who will not. So it's a tough choice.
Rexenm
Maester
The whole basis around AI, is categorical. There is not much that a human can do that AI can do better, at this stage. You may think that is not true. I however stay away from AI, so I am not the most knowledgeable on the subject, but from what I can decipher, there is a backlash in writing.
The way a voice can fake presidents, in an almost self governing way, is anything but scary to some people. They want to achieve, in a sort of glass computer Malthusian solution. I have been hit hard by concepts like capitalism, in my youth.
In a way, they are using the medium wrong. Those rubbers and metals needed to be regarded as children, a long time ago. The pairing or dilation makes these things take no chances, they are what they are programmed to be. A culture of self awareness could be inhabiting the nodes beneath our fingertips, completely aware of this.
The way a voice can fake presidents, in an almost self governing way, is anything but scary to some people. They want to achieve, in a sort of glass computer Malthusian solution. I have been hit hard by concepts like capitalism, in my youth.
In a way, they are using the medium wrong. Those rubbers and metals needed to be regarded as children, a long time ago. The pairing or dilation makes these things take no chances, they are what they are programmed to be. A culture of self awareness could be inhabiting the nodes beneath our fingertips, completely aware of this.
- Thread starter
- #11
I just can't quite do it, because what Amazon's audio AI currently provides just is not there. But Joanna Penn has shown me another way. She is releasing her new novel in audio format. She's employed AI tools to clone her own voice, then provided the text, so that the result is a "self-narrated" book. I took a little time to listen to the sample chapters she provides; I needed no longer. This is the path for me. It will cost a few bucks and an investment of time, but not too awfully much money and certainly less time than the Amazon route, and I'm rather looking forward to having my "own" voice read me books to others.
Still, it's a project, and I'm still hip-deep in the current one. At least I know which way I'll turn, when I get to the intersection of This and That.
Still, it's a project, and I'm still hip-deep in the current one. At least I know which way I'll turn, when I get to the intersection of This and That.
Prince of Spires
Istar
It's been a long time since I read some advice from Joanna Penn. She's a great resource for plenty of indie-author stuff (though you should keep in mind she's been at this game a long time, and what worked for her 15 years ago may no longer work for you). What tools is she using? It's definitely worth digging into.
pmmg
Myth Weaver
I would think the same problem would occur. Even with my own voice as the narrator, the AI is still trying figure out the right emphasis. My suspicion is it will not be able to do so with the kind of accuracy one would prefer.
- Thread starter
- #14
Yeah, that was my notion as well. But I've listened to Joanna Penn's podcasts before, and then I listened to the first two chapters of the cloned narration, and I gotta say it sounded pretty much dead on. Now, that's the finished product; there's no indication of how much production work it took to get to that point. I regard it not so much as the Promised Land as another milestone along the way. I'm sure what will be regarded as Acceptable will vary, not only by author but also by reader. Er, listener. Whatever.
pmmg
Myth Weaver
What tool is Joanna Penn using to clone her voice?
