search

yxlu-0102 / MP-SENet

MP-SENet: A Speech Enhancement Model with Parallel Denoising of Magnitude and Phase Spectra
MIT License
267 stars 40 forks source link

Complex loss calculated using compressed magnitude #32

Closed philgzl closed 1 month ago

philgzl commented 1 month ago

In the long-version paper, the magnitude loss is defined as a distance between the compressed magnitudes (Eq. 2). Meanwhile the complex loss is defined as a distance between the non-compressed complex representations of the signals (Eq. 8). However since the dataset class is returning a complex signal that uses the compressed magnitude here, it seems to me that the complex loss actually uses compressed versions of the signals. Can you confirm?

yxlu-0102 commented 1 month ago

Oops! You are right.

I did use the magnitude-compressed complex spectrum when calculating the loss.

The description in the journal version is incorrect. We will correct it. Thank you for pointing it out!

philgzl commented 1 month ago

Great thanks!