search

Audio-WestlakeU / FullSubNet

PyTorch implementation of "FullSubNet: A Full-Band and Sub-Band Fusion Model for Real-Time Single-Channel Speech Enhancement."
https://fullsubnet.readthedocs.io/en/latest/
MIT License
554 stars 158 forks source link

How to Use Pretrained, pickled model in Releases with No Documentation? #12

Closed uwstudent123 closed 3 years ago

uwstudent123 commented 3 years ago

@haoxiangsnr Hi, it's already April and there still isn't any documentation for the pretrained model in releases. How do we go about using the pickled file data.pkl for inference? Thanks!

haoxiangsnr commented 3 years ago

Hi, Eric.

I'm really sorry. Close to graduation, time was taken up by many things.

I have updated the documentation for the checkpoint usage, and you are welcome to try it.

Please feel free to contact me if you need any further information.

uwstudent123 commented 3 years ago

Ok no worries, thanks for updating! Does the pre-trained model support reverb datasets too? Thanks.

haoxiangsnr commented 3 years ago

Yes. FullSubNet contains a subband-based model, which takes one frequency unit as input each time. This scheme is easy to focus on the difference along the time axis, such as reverberant characteristics. According to the previous experiments, the fullsubnet is more powerful on the "with_reverb" dataset than on the "no_reverb" dataset.

You can try it on your reverb datasets.