Amazon Virtual Voice Audiobooks?

Reader PML writes in to ask:

Several of my favorite offers have opted to use AI Virtual Voice to release some of their older titles in audio format.

I emailed you a long while ago hoping for more audio releases for Caina and Nadia. You indicated that audio publishing is expensive and you preferred to release titles that were not short in length.

I totally understand. But I wondered if you have considered releasing your back titles using virtual voice? The performance is not bad, and I would really enjoy listening to all the books featuring Caina and Nadia. I don’t know what the pricing scale is, but it’s probably quite a bit less than a live reader.

Virtual Voice is Amazon’s new program for creating AI-narrated audiobooks. Will I be using Virtual Voice to turn some of my older titles to audiobook?

No.

Why?

So, there’s three levels to my answer here. 1.) Is it ethical to use AI for audiobook narration? 2.) Is AI narration good enough for audiobook narration? 3.) Does this help visually impaired listeners?

I should mention that I have in fact experimented quite a bit with AI-narrated audiobooks. Part of the reason I did this was because I wanted to understand the technology so I had an informed opinion about it. Google Play beat Amazon to the punch last year, and I experimented with turning the SILENT ORDER series into audiobooks with their technology, since I don’t think SILENT ORDER sells enough to support audiobooks. I didn’t think the AI-generated audiobooks were good enough to sell in good conscience (and just because you’re selling something doesn’t mean anyone will buy it, more on that below), so instead I put the AI narrated audiobooks on YouTube for free. That said, I did turn Adsense on for the audiobooks, so I made a satisfactory if small bit of money from YouTube ads in 2023. You can listen to SILENT ORDER: IRON HAND in AI audio here.

Overall, the response from listeners seemed to be that they loved the story (thanks!) but hated the artificial voice. Like, if they had actually paid for it, I could just imagine the complaints. I think a lot of the authors who create Virtual Voice audiobooks will be disappointed by the response. Like, audiobooks are basically self-publishing on Hard Mode, but if you’re coming to the market with an AI-generated audiobook, it will be even harder to sell than one voiced by a human who knows what he or she is doing.

On to more details!

1.) Is it ethical to use AI for audiobook narration?

“Ethics in AI” is a bottomless quagmire of an Internet discussion. Overall, I think AI technology creates vastly more problems than it solves and is really nothing more than Very Fancy Autocomplete. I also suspect there’s a bit of a bubble to it like there was with cryptocurrency and NFTs. For a while all the Galaxy Brain influencer people thought crypto and NFTs were the future, and then the bubble burst and a significant portion of everything connected to crypto and NFTs turned out to be a big old scam, and all the Galaxy Brains migrated over to touting AI. I suspect a lot of the “AI” technology rushed out now has a same speculative bubble effect, and when the bubble bursts some companies are going to be out billions since they spent all that money building Infinite Crap Generators. Or a lot of people are rushing to shove AI into stuff because it’s trendy, like how the Washington State Lottery decided for whatever reason to put an AI image generator on its site, which it had to pull down hastily when it started generating deepfake nude images.

(It is amusing how some of the really pro-AI Galaxy Brains like to say the US needs to develop AI or else the Chinese will get it first, as if having an Infinite Crap Generator to make deepfake nudes will somehow determine geopolitical dominance in the 21st century.)

But all that said, I don’t think AI is going away. The US courts seem consistent so far in their opinion that AI isn’t plagiarism but it isn’t copyrightable, and there’s a wide range of useful activity in the “not copyrightable but not plagiarism” space. AI can do useful things that crypto and NFTs can’t. Like, suppose you’re applying for forty different jobs, and you could use ChatGPT or Copilot to crank out forty different customized cover letters for your job application. Given how messed-up the job market is at the moment, I could hardly blame someone for doing that. And you see examples of people using generative AI not to create artwork but to handle data processing type chores (like the cover letters) in clever ways that don’t seem cross any moral boundaries.

So I suspect everyone will have to examine their own consciences and decide where their own line is for generative AI.

For me, I decided I’m not going to sell anything that I didn’t make myself (or, in the case of an audiobook, made by a human I hired). If I’m selling something, it was 100% written by Jonathan Moeller, or 100% narrated by a human I hired, and the cover image doesn’t contain any AI-generated art elements. (This is also true of books and stories I give away for free, like my permafree series starters.) That’s where I’ve decided my line is going to be with AI usage.

I have used AI images for Facebook ads, since ads are low-resolution anyway, and you often have to change out the image every week or so anyway. Ad images are often essentially disposable…so why not use the disposable products of AI for them?

2.) Is AI narration good enough for audiobook narration?

All my criticisms of AI aside, “AI voice” or “Virtual Voice” isn’t a new technology. It’s just improved text-to-speech synthesis technology, and text-to-speech has been around since the late 1960s. The AI part just makes the synthetic voice sound closer to an actual human voice than the more obviously artificial tones of older technology. It’s also pretty good at imitating a real human voice, which is why you can go on YouTube and see comedy videos of President Biden trying to make his way through Skyrim or something.

Is the AI narration good enough to support creating a paid audiobook? Well, kinda sorta. It’s good enough now that it creates a near-perfect imitation of a human voice. The trouble is that the voice is so perfect that it triggers the uncanny valley effect where you encounter something that almost seems human but isn’t. It’s also really bad at emotion. The best narrators make it sound like they’re telling a story, and that means varying the emotion of the voice at appropriate times. Text-to-speech simply isn’t very good at that.

That’s part of the reason I won’t use Virtual Voice – I don’t feel the end product is of high-enough quality to sell. Give away for free on YouTube, sure. But sell? No.

It would be good enough for drier non-fiction – like legal casebooks, geological surveys, that kind of thing. A nonfiction book that requires varied emotion – like a war memoir, for instance, or a comedic travelogue – would not work well with AI narration.

3.) Does this help visually impaired listeners?

While I don’t want to use AI narration to create paid audiobooks, I would like to see it become more ubiquitous.

I think the mission of technology is to help us overcome or ameliorate the inherent frailties of the human condition. So I would like to see AI narration eventually become just a button in the ereader app. Like, you hit the read aloud app, and the computer reads to you in a voice of your choosing. You’ll still have the option to buy a human-narrated audiobook, if available, but the option to have the device read to you would be there if you need it.

We’re kind of already there – all the major operation systems for computer and mobile have read-aloud functions, it’s just not implemented consistently and the voices aren’t always very good.

So, I won’t use Virtual Voice or AI narration to create any audiobooks for sale.

In the spirit of full disclosure, as of right now I have agreements with four different narrators to produce four different audiobooks. So I am literally putting my money where my mouth is. 🙂

-JM