shahules786 / mayavoz

Pytorch based speech enhancement toolkit.
MIT License
328 stars 21 forks source link

Wrong sign in imaginary component for ComplexConv2d? #44

Closed mattpitkin closed 1 year ago

mattpitkin commented 1 year ago

In the conv.py file, should lines 73 and 132 be:

imag = real_imag + imag_real

rather than

imag = real_imag - imag_real

This would make them consistent with equation 1 in the DCCRN paper and also consistent with how things are defined in the rnn.py file for the complex LSTM.

izam-mohammed commented 1 year ago

Thanks @mattpitkin for bringing this issue to our attention! We appreciate your contributions to the project. I am currently working on resolving this issue. I will keep you updated on my progress and let you know if there is anything you can do to help. Thanks again for bringing this to our attention.

shahules786 commented 1 year ago

Thanks for you contribution @izam-mohammed :)