Plato's Socrates, Librivox readers and AI.

Everything except LibriVox (yes, this is where knitting gets discussed. Now includes non-LV Volunteers Wanted projects)
Post Reply
dimitris80
Posts: 3
Joined: February 29th, 2024, 10:32 pm

Post by dimitris80 »

Hi.
First post here, hope I posted this in the right forum.

I am creating a free Visual Novel that includes all Plato's works. It is named Plato's Socrates.
An early version without voices can be found here: https://kakoeimon.itch.io/platos-socrates

The texts I used are translations by Benjamin Jowett or George Burges.

I heavily used AI to develop this project. All the graphics are AI generated (It is not that simple as it may sound).
Apart from using AI, my work was to separate the text into speakers and then convert all the books into a simple format that can be used in an engine I wrote for visual novels.

Now I am in the process of creating a voiced version.
All voices were generated by the use of samples from here.
The project is ready, and I am in the proofreading - editing stage, and probably I will release it before I end this stage.

I write this here because I want to acknowledge to the librivox readers that I used their voices.
I hope that I will get their blessings on this and to give credits to whomever wants or have no problem doing so.

Here is my list of readers in their forum names. (I don't know if there is a way to tag them)

Peter Why
pflowerdew
johnnyenglish
redrun
wib66

The project is Open Source and is going to be given for free.
It is going to run to Android, Windows, Linux and Mac.
GrayHouse
Posts: 639
Joined: October 6th, 2012, 3:27 pm

Post by GrayHouse »

I think this is an interesting question that was bound to arise sooner or later. It's an untested legal area, like much of the disruption caused by AI.

Librivox works are dedicated to the public domain. But there may still be legal or moral limits.

If I release a work as public domain, am I surrendering 'likeness rights' in my voice to be used by anyone for any purpose ever? I would like to think that's not the case and that explicit permission should be sought for such activities. Obviously I don't speak for others.
Winnifred
Posts: 2672
Joined: February 4th, 2022, 4:50 pm
Location: North Vancouver, British Columbia, Canada
Contact:

Post by Winnifred »

And beyond the question of “likeness rights,” which I agree should be sought ahead of time, there’s also the question of the quality of the product, especially for Librivox readers, who participate here because they want to provide their authentic voices to the world.

What happens if the AI version of their voice produces dreadful readings? What if the readings are of something they find morally or ethically questionable? Does that risk reputational harm to the individual reader? Some of our readers do this professionally, and could encounter risks to their livelihood because the AI version doesn’t match the standards they set for themselves.

This makes me very nervous.
Winnifred

Readers Wanted:
Where the Blue Begins by Christopher Morley (humorous novel about a "Synthetic Hound" named Haphazard Gissing I.)
Potemkin Village by Fletcher Pratt (science fiction novelet)
redrun
LibriVox Admin Team
Posts: 2941
Joined: August 11th, 2022, 8:32 pm
Contact:

Post by redrun »

In this particular case: this looks like a project I would have liked to be a part of, and I'm glad if my voice is useful to what looks like a free, open-source edu-tainment project!
I'm glad to have heard from you, but I don't need a credit. Folks in my life will know about it only if I choose tell them, which is an arrangement I'm happy with. :lol:

More generally: Some volunteers are glad to see things like this, or to be recruited (though I understand that's not exactly the right term here) for other projects. Others have "done their time" and moved on, and don't really want to be contacted out of the blue.
The way we usually handle this is to move the thread to Off Topic, so I'm going to do the same in this case. :thumbs:

If you're willing to go what might be considered the "extra mile", you might ask if there are other readers who would like their voices to be used, in place of any who don't come along to give their approval. Both the laws and the norms around generative AI are still far from settled, so you've hit upon a subject some of us here have a strong interest in.
I'll be out for a bit on this last weekend of April, but still checking in as I get the chance. I will try to follow up on Monday, with anything I can't do on the go.
annise
LibriVox Admin Team
Posts: 38681
Joined: April 3rd, 2008, 3:55 am
Location: Melbourne,Australia

Post by annise »

Legally as USA law stands, anyone can do anything with our files and we have often had posts from people who use the files in some musical productions they produce.
I think what the first poster is asking is whether the readers would like to be credited or not.
So that is up to the people involved. What he has done is quite legal under current laws.

Anne
dimitris80
Posts: 3
Joined: February 29th, 2024, 10:32 pm

Post by dimitris80 »

I am not an expert in AI, but just a user of the technology.
In this case, I use Tortoise-tts : https://github.com/neonbjb/tortoise-tts
But as far as I understand, you need thousands of hours of audio to train an AI model, you also need to transcribe the audio and crop out every external noises or sounds and to be sure that all the audio is from one language without accents. For example, English but without heavy British accent.
Then you need a farm of GPUs to run for several days.
In the end, you will get a model that will be able to output English and probably other languages that belong to the same language family.
With this model now (in the case of Tortoise-tts) you need at least 3 samples of the voice you want to clone with a duration of around 10 secs each.
You do not even need the voice samples to be in the target language (English) or accent.

There are also tools to detect if an audio was generated by AI. The author of tortoise-tts wrote one for his tool.

In the end, everyone who ever gave an interview, read a book online or made a podcast etc. is exposed.

It is not just people who made a reading like in this site, but everyone who had a minute of "fame" and his voice was clean recorded.

But I am not going to lie, the number one stop at least for me was librivox and the reason was the permissive licence.

Now let's get into the Dark side.

DeepFakes are here for far too long. I have seen people (mostly women) with psychological meltdown. The reason that are mostly women is that they are the targets of the AI users and nothing else. Everyone is vulnerable.
But on the other hand all the deepfake videos were exposed.

In my opinion, the problem is not the damage to fame or integrity of the targeted individual. But the psychological damage that can be done, when you see or hear yourself in a very inappropriate action. It sounds quite simple to the observers, but I have never seen anyone to take it easy when he is the target.
dimitris80
Posts: 3
Joined: February 29th, 2024, 10:32 pm

Post by dimitris80 »

The moment of truth is here.
The project is up for windows for the time being.

Here is the link: https://kakoeimon.itch.io/platos-socrates

Download the file plato_s_socrates_talkie_0.1.exe

In the end I gave credits only to the site (LibriVox) I haven't mentioned anyone, in sort is like I followed redrun's preference.

Credits can be given and in later versions as this is the one something like beta.

Here it is and a twitter post I made for this with a video.

https://x.com/kakoeimon/status/1767863359704015333?s=20

Hope you all enjoy it.
Post Reply