where did you get the model for melspectrogram.onnx?

[skewballfox]

is it being generated in the code? if so, could you point out where? I see it being loaded in openwakeword/utils.py but not generated. The reason I'm asking is because loading it into netron, it seems to be using a older onnx op set, and I wanted to see if making use of operators added to onnx after 1.13 like STFT could make a difference in speed or model size.

[dscripka]

There is an example notebook here that shows how the melspectrogram model is defined and saved to an ONNX file.

I haven't tried it with new operators, but it's possible that it could be made faster or smaller with some experimentation.