[REQ] OpenVINO backend too

[MarcoRavich]

Hi, first of all thanks for this great work!

As @WolframRhodium and @AkarinVS have enabled in their vs-mlrt project, it would be cool to let Intel-hardware owners to take advantage of anira too.

You may also ask @RyanMetcalfeInt8 - who maintain OpenVINO™ AI plugins for Audacity - of technical aspects about.

Hope that inspires !

[RyanMetcalfeInt8]

Thanks @MarcoRavich!

My quick thoughs -- looks like as there is already an onnxruntime backend for this project, Intel-hardware optimizations could be enabled by making use of the OpenVINO EP for onnxruntime (https://skottmckay.github.io/onnxruntime/docs/reference/execution-providers/OpenVINO-ExecutionProvider.html).

[MarcoRavich]

Thanks @MarcoRavich!

Oh, yup, would love to see "OpenVINO™ VST3/JUCE Audio AI plugins" too ! ;)

[faressc]

Hi @MarcoRavich, hi @RyanMetcalfeInt8,

@MarcoRavich thanks for the kind words and interest in the project! As @RyanMetcalfeInt8 pointed out, we already have an ONNXRuntime backend that could potentially use the OpenVINO™ Execution Provider. But, in order to be compatible with a wide range of CPUs, we do not enable it by default.

However, you can take advantage of anira's ability to use a custom backend. To do this, you can pass a custom BackendBase implementation to the InferenceHandler. Essentially, you would copy anira's existing OnnxRuntimeProcessor implementation and adapt it to use the OpenVINO™ Execution Provider.

Another option would be to use a native OpenVINO implementation within the custom backend, as is the case with the OpenVINO™ AI plugins for Audacity.

For more details, check out our usage guide for a custom backend.

Cheers and let me know if I can help you further!