Is there a way to access the separation mask?

[hananbo26]

When I am using a pretrained model to infer.
Is there a way to access the separation mask in the STFT domain?

[mpariente]

Have a look at the functions in BaseModel.forward, there are function calls and it's pretty clear where is the mask.

[hananbo26]

Thanks for your response🙏

I succeeded to access the TF masks estimated by a pretrained model but to the best of my understanding these masks are not in the STFT domain but in some other TF representation.

I need the mask in the STFT domain.

Is there a pretrained model for speech separation that uses the STFT as encoder? or any other way to access the STFT masks?

[mpariente]

Great!
I don't think we have any STFT-based pretrained model for now, sorry.
If you do train one, please share it 😉