Gutenberg Audiobooks

[msfry]

I was reading through Gutenberg's Computer Generated Audiobooks pages today, which got delivered to my phone via a Facebook page, and here are two screenshots of interest. The first names Librivox as a contributor of AI books, which I think is wrong and should be corrected. The other talks about computer generated audiobooks taking only 30 seconds to create in mp3 format. Can that be?

https://1drv.ms/i/s!AqmYAdtBQdx8itJvAsoZviNcONY39Q?e=SOyaqT
https://1drv.ms/i/s!AqmYAdtBQdx8itJwS8FEml9xXFR3VA?e=pSzWzb

Has anyone ever listened to one of these audiobooks. How is it done? And what's their process?

[TriciaG]

Rather than a contributor, it says we're a recipient/distributor of their AI-generated audiobooks, which is not true (of course). If there's a contact link somewhere, it would be good to use it and ask them to remove our name from that page.

I have no comment about how long it takes, nor have I listened to any of them.

[Availle]

They may mean the LibriVox app (which has books from different sources) and not our catalog though.
Still, some clarification would be appreciated.

[msfry]

They may mean the LibriVox app (which has books from different sources) and not our catalog though.
Still, some clarification would be appreciated.

I'll try to get that clarified and report back.

[msfry]

I sent Gutenberg this email:
Per the attached screenshot from my iphone, received yesterday from I think a Facebook page, I would like you to correct the error that lists Librivox as a distributor of AI audiobooks. You may be referring to the Librivox Audiobooks App, which has no affiliation or connection with Librivox, other than that they freely distribute many of our audiobooks as well. I thank you in advance for attending to this matter, and would appreciate hearing from you when it is resolved.

Michele Fry, Librivox Volunteer

Meanwhile, here's the Home Page link, and the Project Gutenberg Audiobook Collection, showing 4840 titles to date.

I picked one at random from the first row, Lost Pond, by Henry Abbott, 24 minutes long, just to experience it. It begins with a robot lady's introduction, then the story begins with a man's voice which sounds just like an old friend of mine. It's quite listenable, not at all robotic.
Speed is variable in .25% increments, and the other controls seem the same as for our audiobooks.
The Chapters are not listed out into sections, as are ours, and I didn't notice the ability to bookmark on IA, but maybe it is possible on the other podcast players.
There is a Feedback Survey for any book you experience (which being as there are no readers' feelings to hurt, that's helpful.)

[lightcrystal]

I've listened to a few. Are they using the same AI voice for them all? An American male voice. Don't ask me which state it is. It's clearly not Queensland. [that's a joke].

[msfry]

Are they using the same AI voice for them all?

Your question piqued my curiosity, so I sampled a random assortment. Here's what I found:

• The 6 I chose all appear to be in that same male voice!

• The book with the most views, Pride and Prejudice 320 views, yep, same voice.

• Next was Power of Concentration, which weirdly begins with a very long disclaimer, same voice, of how you may and may not use the etext material, with reading of punctuation marks (asterisk, asterisk, forward slash) and pacing way off.

• Unlike the book I listened to yesterday, on another one the AI did not comprehend what it was reading, so there was no nuance or pacing.

• Then came one with a reading of the Table of Contents before commencing with the actual book text.

• Another book began by reading the Title Page word for word, which included other books by the same author, illustrator, publisher, and so on.

Not an exhaustive sample, but it leads me to believe they are still just toying around with the idea, and no Proof Listening oversight.

[GettingTooOld]

The other talks about computer generated audiobooks taking only 30 seconds to create in mp3 format. Can that be?

It's a trivial matter even on a modern domestic desktop. Speech is not that complex and it's really not that fast as far as computing power keeping up with it on the fly so to speak. I guess for comparison, it's a challenge for some computers to keep up with analyzing or synthesizing radio frequencies and coding and decoding information in those.

https://1drv.ms/i/s!AqmYAdtBQdx8itJvAsoZviNcONY39Q?e=SOyaqT
https://1drv.ms/i/s!AqmYAdtBQdx8itJwS8FEml9xXFR3VA?e=pSzWzb

These don't work on my computer, the ms.drv site is at fault.

Has anyone ever listened to one of these audiobooks. How is it done? And what's their process?

https://api.products.aspose.app/ocr/en/image-to-speech

This site I found for another thread yesterday, and it not only creates the voice on the fly, it also takes a picture of a page as input, or a screenshot. So you could load up any page of text, not just gutenberg, and it will convert the image to text and then to an audiobook for you. It took a minute or something like that for it to spring up for me, I wasn't watching. The whole thing is perhaps poor quality and makes errors, but it was the first cab off the rank I found to help someone who can't see, so they'll either adapt to it as blind people always do in my experience or find a better one.

Soon or perhaps already, I've been trying to find a freeware version, you should be able to do it completely offline on an old laptop. You should be able to take a minute long sample of the voice of your loved one who has departed, a current love, or a celebrity you like, or hate and load that in, and then your laptop will read whatever you like using the voice you want. That's pretty good I'd say.

Simple text to speech is decades old already, even on freeware operating systems. It's simply getting better as far as quality is concerned. I can always tell the difference so far, it will take an artist engineer and magician combined before something comes out where you can't tell the difference from real. Maybe 5 years online, maybe 7 years for offline, and ergh, maybe a dozen or more for it to be freeware.

[GettingTooOld]

I listened to lost pond, that's good quality, it has a supercomputer(cloud type) behind it. Plus, unfortunately, it's copyright is questionable, as no-one has yet decided about AI performances and who owns them, whatever they eventually do come up with will probably be retrospective and change here and there. You could upload it yourself anywhere and even sell it, but that may all change later on. So it can't be used in a lot of ways, like distribution on freeware platforms and the like, where it needs to be definite.

[msfry]

I ran across both present and read mispronounced yesterday in an online article. Pleasant voice or no, AI can't tell the difference, so pronounces them both the same:
I present you with a present.
Please read the book I read you earlier.

[msfry]

https://api.products.aspose.app/ocr/en/image-to-speech

I looked at the aspose app page. I can see its value in reading contracts, ads, recipes, legal briefs and other documents aloud to anyone as they're driving down the road or cooking dinner. Who has time to sit and read. Taka a picture, and listen. But of course you'll need a cameraphone and/or scanner, and not mind the lack of expression and mis-pronunciations. That complaint notwithstanding, AI stocks have exploded these past 6 months, so there's no doubt people believe in them.

[GettingTooOld]

But of course you'll need a cameraphone and/or scanner, and not mind the lack of expression and mis-pronunciations.

Don't forget screenshots ! I forgot them at first, but screenshots from all around the internet, all around your computer, it all works too.

[lightcrystal]

It's worth adding here that I am narrating a book that is impossible to do with AI. If you think that Microsoft AI could handle Christian Astrology then you are not familiar with 1600s publishing standards.
Microsoft can't do it!
But I can I can I can...