4 ms·

I hope AI voice tech like https://beta.elevenlabs.io/ https://beta.elevenlabs.io/ will make audiobooks even better.

by kingstoned 3y ago

chrisco255 3y ago
It definitely seems like it should be plausible to offer multiple AI readings of all public domain books.
- prepend 3y ago
  It seems like it should be plausible to offer AI readings of any book I’ve purchased. It seems strange that ebook readers aren’t allowed to run the ebook I’ve purchased through text to speech.
em-bee 3y ago
i am waiting for the day where i can use my own computer to generate high quality reading of any text so that i can listen to any book i like without waiting for an audiobook to be published.
- Micheal_McL 3y ago
  Same. Feel uneasy having audio books on Audible or Apple Books. They have the power to remove them or block your account at anytime and the books are no longer in your possession. Highly unlikely ofc.
- sandreas 3y ago
  It's already possible, I'm working on the exact same thing. Take a look at https://www.thorsten-voice.de/en/motivation-vision-english/ https://www.thorsten-voice.de/en/motivation-vision-english/. He is a german guy who is donating his voice to the community and provides a LJSpeech dataset (https://www.thorsten-voice.de/en/datasets-2/ https://www.thorsten-voice.de/en/datasets-2/) as well as lot's of information how to use it (the quality is not as good as the commercial ones, it really is usable). While his voice is not that bad, I would prefer my favorite german narrators to read my audiobooks, so I'm working on creating custom datasets from existing audiobooks that I own. These of course cannot be published because it would violate terms, but you can use ffmpeg to split an existing audio book by silence and transcribe it automatically using existing tools to create your own LJSpeech dataset. There are also LJSpeech datasets for other languages - the format is pretty simple and can be used by anyone to train AI models. The result is pretty impressive for "offline only".
  - em-bee 3y ago
    that is interesting! can you describe how that transcription process works? if i have an audiobook and the corresponding ebook/pdf, isn't that already transcribed? or does transcription here mean something else? i'd also be happy to use an existing voice (the english one from keith ito sounds pleasant enough) but i am confused how to use it to read a book. there is code for a model that learns to synthesize speech from the data: https://github.com/keithito/tacotron https://github.com/keithito/tacotron but i don't see how to get at the end result which i think would hopefully also be available somewhere so i can just use it to read something.

I hope AI voice tech like https://beta.elevenlabs.io/ https://beta.elevenlabs.io/ will make audiobooks even better.

chrisco255 3y ago

It definitely seems like it should be plausible to offer multiple AI readings of all public domain books.

prepend 3y ago
It seems like it should be plausible to offer AI readings of any book I’ve purchased. It seems strange that ebook readers aren’t allowed to run the ebook I’ve purchased through text to speech.

em-bee 3y ago

i am waiting for the day where i can use my own computer to generate high quality reading of any text so that i can listen to any book i like without waiting for an audiobook to be published.

Micheal_McL 3y ago
Same. Feel uneasy having audio books on Audible or Apple Books. They have the power to remove them or block your account at anytime and the books are no longer in your possession. Highly unlikely ofc.
sandreas 3y ago
It's already possible, I'm working on the exact same thing. Take a look at https://www.thorsten-voice.de/en/motivation-vision-english/ https://www.thorsten-voice.de/en/motivation-vision-english/. He is a german guy who is donating his voice to the community and provides a LJSpeech dataset (https://www.thorsten-voice.de/en/datasets-2/ https://www.thorsten-voice.de/en/datasets-2/) as well as lot's of information how to use it (the quality is not as good as the commercial ones, it really is usable). While his voice is not that bad, I would prefer my favorite german narrators to read my audiobooks, so I'm working on creating custom datasets from existing audiobooks that I own. These of course cannot be published because it would violate terms, but you can use ffmpeg to split an existing audio book by silence and transcribe it automatically using existing tools to create your own LJSpeech dataset. There are also LJSpeech datasets for other languages - the format is pretty simple and can be used by anyone to train AI models. The result is pretty impressive for "offline only".
- em-bee 3y ago
  that is interesting! can you describe how that transcription process works? if i have an audiobook and the corresponding ebook/pdf, isn't that already transcribed? or does transcription here mean something else? i'd also be happy to use an existing voice (the english one from keith ito sounds pleasant enough) but i am confused how to use it to read a book. there is code for a model that learns to synthesize speech from the data: https://github.com/keithito/tacotron https://github.com/keithito/tacotron but i don't see how to get at the end result which i think would hopefully also be available somewhere so i can just use it to read something.