Onnx model
Hi, thank you for amazing work! I would like to use the model with onnxruntime, so I won't need to install heavy dependencies. How can I use it with that?
That would be very useful. Especially for running it on different machines. Onnx also has webgpu integration recently, which is pretty cool for a cross-platform experience. :-)
I've been looking into the codebase but its obvious quite complex. core TTS likely involves the T3 and S3Gen (sep models). For an ONNX conversion, which would likely require model splitting, I'd be grateful for any guidance from the maintainers, particularly on strategies for handling the dynamic control flow of the AlignmentStreamAnalyzer, the iterative ODE solving loop within S3Gen's ConditionalCFM, and any custom operations like SineGen or the Snake activations. Any pointers on how these could be made ONNX-compatible would be a fantastic
Thanks for this great work!
Very interested in this too! 🤗
Edit: @VladOS95-cyber and I got it working!
-
English-only: https://huggingface.co/onnx-community/chatterbox-ONNX
-
Multilingual: https://huggingface.co/onnx-community/chatterbox-multilingual-ONNX
Enjoy!
+1
I privately attempted an ONNX migration. I was able to get each of the models ported, apart from the VO Encoder.
+1
I privately attempted an ONNX migration. I was able to get each of the models ported, apart from the VO Encoder.
@nickheyer hey buddy i am eagerly looking for onnx option for chatterbox. it would be really great if you could share your findings. thanks in advance
+1
+1
For anyone still needing the ONNX model: https://huggingface.co/onnx-community/chatterbox-multilingual-ONNX
Non-multilingual Model: https://huggingface.co/onnx-community/chatterbox-ONNX
They seem to work.