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

, I would like to ask the four folders were placed what is the document?

Open 867923756 opened this issue 7 years ago • 4 comments

Thank you for sharing, I would like to ask the four folders were placed what is the document? Can you explain it in detail? noise_dir, speech_dir, val_noise_dir, val_speech_dir

867923756 avatar Jan 31 '18 07:01 867923756

Glad to know you like it. They are directorys for speech and noise used during training and cross validation.

zhr1201 avatar Feb 01 '18 06:02 zhr1201

I just started to do speech enhancement, very interested in this area, very pleased to be able to get your help! And，What kind of noise is mainly used to deal with this project? How much data do I need to train my model?

867923756 avatar Feb 01 '18 06:02 867923756

There is a free dataset online with 100 types of noise (sorry I forget its name), and it can be used to test the efficacy of CNN for enhancement. A larger free dataset called AudioSet by Google can also be used here.

zhr1201 avatar Feb 01 '18 07:02 zhr1201

Thanks for your sharing,but I am confused that why validation_samples = 848824? I download the TIMIT data set, for train set are 4680 waves,for test set are 1620 waves,for Nonspeech are 100 types of noise,could you please help me to figure out why validation samples are 848824? Thank you very much!

deeplearningzhy avatar Dec 04 '18 13:12 deeplearningzhy