search

jzi040941 / PercepNet

Unofficial implementation of PercepNet: A Perceptually-Motivated Approach for Low-Complexity, Real-Time Enhancement of Fullband Speech
BSD 3-Clause "New" or "Revised" License
333 stars 92 forks source link

RSS+NS based on PercepNet #52

Open cloudvc opened 2 years ago

cloudvc commented 2 years ago

Hi all,

Hope to discuss the RSS+NS based on PercepNet. I want to combine the AEC linear part from webrtc with PercepNet implementation to improve the RSS/NS of webrtc.

jzi040941 commented 2 years ago

Hi, that's interesting thing, I'm happy to extend percepnet, author of PercepNet already implemented it in this papaer https://arxiv.org/pdf/2102.05245.pdf probably you should follow this approach.

the main different between percepnet and aec-combined-percepnet is percepnet => input = noisy , output = clean, aec-combined-percepnet input = (aec-processed-noisy, farend signal) ,output = clean

to do so you should make framework which can generate aec processed noisy signal, from far-end and noisy file also, you need to add another input head to percepnet model

however, on a recent paper MTFAANET, they use aec-processed noisy singal as conditional input and they apply mask to original noisy file to reduce distortion from aec.

cloudvc commented 2 years ago

Thanks for your guide. I will follow MTFAANET too.