Audiobooks & AI

[Dopedangel]

I personally don't listen to Audio books as for some reason I am never able to concentrate properly. But I saw the demo of this LLM model and how good it sounds I thought I would share.
https://yummy-fir-7a4.notion.site/dia

[jonathan2022]

Are the voices that Microsoft edge use considered AI ? They aren't bad just wish it had an offline option.

[jonathan2022]

Personally I would have preferred they just kept the AI as a tts option and the real voice actors for paid audiobooks.

[jonathan2022]

And when I say keep AI i don't mean current voiceview voice because that has gotten robotic.my favorite tts currently to use are Google tts on android and libritts_r option on calibre.

[arjaybe]

I spent four-and-a-half years recording 3500 phrases for an outfit called VocalID so they could use my voice to give someone a natural sounding voice when they have lost theirs. Now AI can do it in less than 5 minutes.

I did get a certificate, though.-)

Also, if AI could replicate the way I record stories, I might let it do it for me. My own voice sounding natural? I wouldn't mind saving the time.

[jonathan2022]

I could be wrong, but the piper tts project has tools that you can use to set that up. Making tts with your own voice. Never tried to train with my own voice so don't know what it's like.

[jonathan2022]

Thanks arjaybe I didn't know about this project.

[Quoth]

Originally Posted by arjaybe

Also, if AI could replicate the way I record stories, I might let it do it for me. My own voice sounding natural? I wouldn't mind saving the time.

It can sound like you, but not manage how you would read a book. No intelligence, i.e. no understanding of context or emotion.

[salamanderjuice]

Originally Posted by Quoth

It can sound like you, but not manage how you would read a book. No intelligence, i.e. no understanding of context or emotion.

There are models that can be annotated with emotions.

Does AI surpass the best human readers? No. But they are leagues above the old TTS and honestly a lot of people if my high school English class was remotely typical.

All I want is them to sound natural and not flub words.

IMO I could listen to whole book if it sounded like this.

[Quoth]

Originally Posted by salamanderjuice

There are models that can be annotated with emotions.

Does AI surpass the best human readers? No. But they are leagues above the old TTS and honestly a lot of people if my high school English class was remotely typical.

All I want is them to sound natural and not flub words.

IMO I could listen to whole book if it sounded like this.

There are none that can read text properly without extra contextual commands put by humans. It's all a scam (AI). It just sounds better in demos. I was invited and used Google's bleeding edge to a book. I did just one chapter and while it sounded nice, it was a failure. You could script intonation, emotion, loudness and phonetic hints in the 1980s.

It's even worse if it's not a USA English text. In other news, AI powered self-service supermarket checkouts train humans to steal, but since it's cheaper than the Irish minimum wage including losses they don't care.

[salamanderjuice]

Originally Posted by Quoth

There are none that can read text properly without extra contextual commands put by humans. It's all a scam (AI). It just sounds better in demos. I was invited and used Google's bleeding edge to a book. I did just one chapter and while it sounded nice, it was a failure. You could script intonation, emotion, loudness and phonetic hints in the 1980s.

It's even worse if it's not a USA English text. In other news, AI powered self-service supermarket checkouts train humans to steal, but since it's cheaper than the Irish minimum wage including losses they don't care.

Did you listen to example I linked? I didn't even script any specific intonation, emotion or loudness or phonetic hints.

This is the entirety of what I fed parler-TTS to make that snippet:

prompt =  "A long line of light still lingered behind the purple hills upon which there stood out a few houses silhouetted against the primrose yellow of the sky. Within the nearer confines of the lawn the leafless trees looked dim and shadowy. The whirl of eddying leaves against the wall sounded ghostly, and the rattling of the vines against the side of the house suggested wintriness."

description = "Laura clamly reads a book to her child. The recording is of very high quality, with the speaker's voice sounding clear and very close up."

The description is specifying a specific voice and not much else.

I've heard much worse from real people.

[Quoth]

In the real world you can't produce a decent audio book by computer and sell it. It's tantamount to fraud. Buy the ebook and try listening to a whole book.

That demo isn't enough.

[Quoth]

Originally Posted by salamanderjuice

I've heard much worse from real people.

They probably don't get employed to narrate audio books, or as news presenters.

[arjaybe]

Originally Posted by jonathan2022

I could be wrong, but the piper tts project has tools that you can use to set that up. Making tts with your own voice. Never tried to train with my own voice so don't know what it's like.

Thanks, Jonathan. I'll explore piper tts and see if it's a fit for my system.

[arjaybe]

Originally Posted by Quoth

It can sound like you, but not manage how you would read a book. No intelligence, i.e. no understanding of context or emotion.

Yes. I would need it to do my pacing and inflections and I don't see how it can do that yet. Maybe if it listens to the ten novels I've recorded so far?