rocca

/

lyra-v2-soundstream

like 6

Hey @roseman , yep I linked the tflite-to-onnx conversion script in the readme of that repo: https://colab.research.google.com/gist/josephrocca/401efe82bf18ffe93d5f3deca7ed7515/notebook.ipynb]

I'm currently blocked on this issue for the encoder: https://github.com/onnx/tensorflow-onnx/issues/2059#issuecomment-1296301499

Please share your progress in getting it working on .NET (either here or in the github issues), since it may be helpful to others.

Thanks for the reply. I have the same issue with the encoder. Strange.

Inference with latest 1.3 files is OK now. However when reconstructing audio I get noise only. Maybe I'm missing something.

@roseman Some things to take note of:

• https://github.com/onnx/tensorflow-onnx/issues/2059#issuecomment-1317139150
• https://github.com/microsoft/onnxruntime/issues/13670#issuecomment-1320309501

Can you create a simple Github repo like https://github.com/josephrocca/lyra-v2-soundstream-web with a Colab notebook and inference comparisons? You can compare against TFLite inference numbers here: https://colab.research.google.com/gist/josephrocca/ba3021195dfb4f91ffc7898cea6f8596