How do I choose the embedding model for speech diarization?

[takipipo]

Wow, your work is incredible! 👏🏻

I’m curious about how to pick the embedding model from the ones listed here: https://github.com/k2-fsa/sherpa-onnx/releases/tag/speaker-recongition-models. There are a lot of options to choose from.

I tried to find the original model by looking at the file names, but I couldn’t find and not so sure that the one I found is equivalent to the one you exported to onnx. Is there a link that shows which eager model is compiled to ONNX? That way, I can check out their paper and see if it’s what I’m looking for.

[csukuangfj]

but I couldn’t find

Please describe in detail what you have done to find the original models.

[csukuangfj]

Also, please answer

Please describe in detail what you have done to find the original models.

[takipipo]

Also, please answer

Please describe in detail what you have done to find the original models.

google :/

[takipipo]

Please give an example about the filename that you think is confusing.

Do you have the naming format for the model file like
wespeaker_en_voxceleb_resnet34.onnx is <framework>_<lang>_<dataset>_<backbone>.onnx

[csukuangfj]

wespeaker_en_voxceleb_resnet34.onnx

We use framework_modelName.

wespeaker_en_voxceleb_resnet34.onnx:

• wespeaker, is the framework.
• en_voxceleb_resnet34.onnx is the model from the wespeaker framework.

You need to ask the developers of wespeaker for the meaning of en_voxceleb_resnet34.onnx by yourself.