'SimAMResNet34' is not working properly.

[NathanJHLee]

Hi, wespeaker team.

You have released SimAMRestnet model, So would like to check DER on Speaker Diarization.
But I got error about 'invalid dimensions for input'.
I think if f-bank extracts feature and gives it to model, your model is ready to take only 200 frames.
I also checked voxceleb_resnet34.onnx works fine.

Error is here.

(wespeaker) asr@kpc:~/workspace/wespeaker/examples/voxceleb/v2$ python wespeaker/bin/infer_onnx.py --onnx_path ./voxblink2_samresnet34.onnx --wav_path /ws/stt/DB/SD/wespeaker/voxconverse_data/test/voxconverse_test_wav/aepyx.wav
Traceback (most recent call last):
File "/home/asr/workspace/wespeaker/examples/voxceleb/v2/wespeaker/bin/infer_onnx.py", line 72, in
main()
File "/home/asr/workspace/wespeaker/examples/voxceleb/v2/wespeaker/bin/infer_onnx.py", line 66, in main
embeddings = session.run(output_names=['embs'],
File "/home/asr/miniconda3/envs/wespeaker/lib/python3.9/site-packages/onnxruntime/capi/onnxruntime_inference_collection.py", line 220, in run
return self._sess.run(output_names, input_feed, run_options)
onnxruntime.capi.onnxruntime_pybind11_state.InvalidArgument: [ONNXRuntimeError] : 2 : INVALID_ARGUMENT : Got invalid dimensions for input: feats for the following indices
index: 1 Got: 16913 Expected: 200
Please fix either the inputs/outputs or the model.

(wespeaker) asr@kpc:~/workspace/wespeaker/examples/voxceleb/v2$ python wespeaker/bin/infer_onnx.py --onnx_path ./voxceleb_resnet34.onnx --wav_path /ws/stt/DB/SD/wespeaker/voxconverse_data/test/voxconverse_test_wav/aepyx.wav
(1, 256)

[wsstriving]

@cdliang11 @JiJiJiang Hi Chendong and Hongji, can you help to check whether it's the problem of the onnx export? Thanks

[cdliang11]

@cdliang11 @JiJiJiang Hi Chendong and Hongji, can you help to check whether it's the problem of the onnx export? Thanks

voxblink2_samresnet34.onnx model has a fixed input shape, so we need to re-export an onnx model with dynamic input.