Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add kokoro tts #11

Open
thewh1teagle opened this issue Jan 5, 2025 · 9 comments
Open

Add kokoro tts #11

thewh1teagle opened this issue Jan 5, 2025 · 9 comments

Comments

@thewh1teagle
Copy link

Very cool project!
Consider use https://github.com/thewh1teagle/kokoro-onnx
It supports raspberry pi

@OriNachum
Copy link
Owner

I used piperTTS.
Is that a good alternative, or is kokoro superior / more fitting?

@thewh1teagle
Copy link
Author

I used piperTTS.
Is that a good alternative, or is kokoro superior / more fitting?

The quality is so much better.
It's a bit heavier. on macOS it works 4x faster (M1 CPU) on Windows 2x faster (Ryzen 5 NPU).
On Raspberry PI 4, it was much slower.
But I believe that if you have NPU/GPU onnxruntime will take advantage of it

@OriNachum
Copy link
Owner

I am integrating Nvidia Jetson Orin Nano Super 8GB.

It’s worth adding support if only for the research and POC.

@OriNachum
Copy link
Owner

First, I saw your post on Reddit
Second, I saw how easy it is to add, and I did most of the work on my Jetson Orin Nano.

Going to implement for Orin Nano tonight.

Kudos for the work - I would love to give acknowledgment and link your repo after I integrate and demo it.

@OriNachum
Copy link
Owner

@thewh1teagle , testing on Nvidia Jetson Orin Nano Super 8GB.

No TensorRT support, no CUDA support or it is inefficient.

I can't check CPU or CUDA specifically, as working with session requires EspeakConfig.
But stream works great!

It does feel I'm choking resources on the Jetson, with thinking and speaking at once.

Should I open an issue in your repo?

@thewh1teagle
Copy link
Author

as working with session requires EspeakConfig.

It shouldn't be related

No TensorRT support, no CUDA support or it is inefficient.

Cuda supported
Try to enable the executionProvider

Should I open an issue in your repo?

Yes that can be helpful anyway

@OriNachum
Copy link
Owner

I'll give it another look and improve logging.

Thank you for the quick replies and reaching out!

@OriNachum
Copy link
Owner

OriNachum commented Jan 11, 2025

@thewh1teagle with session I can force use CUDA
It runs at 643mb and took 0.47seconds for "Hello world"

I needed to pass "espeak_config=None" on create_with_session.

It looks like a trivial/minor bug - I'll open an issue

Also, I'm going to add this to my demo and guide before I publish it on medium and relevant groups.

@OriNachum
Copy link
Owner

Just pushed an example class to the demo app.
Keeping this open for the refactor on the main project

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants