CNN-for-single-channel-speech-enhancement icon indicating copy to clipboard operation
CNN-for-single-channel-speech-enhancement copied to clipboard

Published 20 hours ago • verified publisher icon zhr1201

Request to upload a pretrained model

Open absurdvision opened this issue 7 years ago • 4 comments

Hi Would it be possible to upload a pretrained model. It would be very helpful to have a starting/reference model to further tune the network.

absurdvision avatar Feb 03 '18 05:02 absurdvision

Sorry but it exceeds GitHub's file size limit of 100.00 MB.

zhr1201 avatar Feb 03 '18 06:02 zhr1201

Would you be able to upload to Google drive, Dropbox or something similar and put a link in readme ? It would be very helpful for users without a big GPU

absurdvision avatar Feb 03 '18 06:02 absurdvision

Thank you for your great advise. I am now not able to access my labs' server because it's Winter Vocation in China and I am not at my school. After the Spring Festival, I will upload my model.

If you just have like less than 100 types of noise, I believe training the current model with Adam and tuning the learning rate a little bit will work fine. If you are dealing with a larger noise dataset, I believe you should probably adding more layers and using skip connections.

zhr1201 avatar Feb 03 '18 06:02 zhr1201

Hi, I would also be interested in a trained model.

anicolson avatar Jan 25 '19 03:01 anicolson