Integrate Kokoro-onnx TTS

[aedocw]

I just played around with https://github.com/thewh1teagle/kokoro-onnx and it's pretty amazing. The voices sound great, and it's really fast. I'm going to incorporate this as an engine option either here or with epub2tts-edge (or maybe just do a brand new epub2tts-kokoro).

[zos474]

I'd vote for this package. epub2tts is the best package I have come across for realistic audio creation, with epub2tts-edge, the second (and, believe me, I have wasted countless frustrating hours trying to get many others to work, to say nothing of probably a 100 GB of downloads along the way... And when I got them to work, the quality and 'naturalness' of epub2tts is head and shoulders above anything else.) So, major kudos to you on these packages! What I prefer about epub2tts vs edge version is that the quality/cadence is better (to my ears) and I absolutely love that you have whisper do a check of the output. And the voice cloning just works. Thanks again for all your work on this.

[aedocw]

Thank you so much for the kind words @zos474 , I really appreciate it!

I played around with kokoro last night and made some progress. The main reason I started to play with the -edge version is just because it's a simpler code base. It also chunks into paragraphs more cleanly, so I thought it would be a good test.
I need to do some more experiments but I've hit some errors and even crashed a machine (but that could have been the GPU getting oversubscribed from something else). Once I have better error handling as well as an idea of how much text you can send at once, I'll poke at adding it as an optional engine to epub2tts.

It should be pretty straightforward except for downloading the model and voices.json. Actually that won't be difficult, mostly just saying that out loud here so I remember to think about a clean way to do it.

[aedocw]

WOW, gotta say this implementation is so good, I would suggest anyone who wants to make an audiobook with Kokoro just use this guys project:

https://github.com/santinic/audiblez/

Personally that's what I'll be using to make audiobooks at this point. Kokoro is hands down the best sounding TTS (IMO), and this project is clean, and works beautifully. To be honest it puts my stuff to shame hahaha!

[aedocw]

I do have a branch that sort of works with kokoro anyway, if anyone wants to play with it.

https://github.com/aedocw/epub2tts/tree/add-kokoro

[aedocw]

GUI for aidiobook creation with kokoro https://github.com/plusuncold/autiobooks

[Nikanoru]

Do you know if this also works with .txt files or does it require the files to be in epub format?

[aedocw]

Looks like it only takes epub.

I have a version of epub2tts-edge that I've been using that does kokoro instead of edge, and has worked great (and works with text files). It was really easy to use that one because the requirements are minimal compared to this project (because this project supports so many different TTS engines, it requires all of their libraries to be installed even if you only want to use one).

I'm going to be pretty busy for the new few weeks but if I get some spare time I'll clean it up a little and put it up on github and will link here.

[Nikanoru]

Looks like it only takes epub.

I have a version of epub2tts-edge that I've been using that does kokoro instead of edge, and has worked great (and works with text files). It was really easy to use that one because the requirements are minimal compared to this project (because this project supports so many different TTS engines, it requires all of their libraries to be installed even if you only want to use one).

I'm going to be pretty busy for the new few weeks but if I get some spare time I'll clean it up a little and put it up on github and will link here.

That sounds fantastic. Your implementation is what I enjoy the most, and this Kokoro engine really sounds great.
I still hope there will be a way to donate money to you for your work some day. :)