spxnn / SDNet

Pytorch implemention of SDNet

SDNet:Speaker and Direction Inferred Dual-channel Speech Separation

Requirements:

Pytorch>=1.1.0
resampy
soundfile

Model Descriptions:

Data Preparation

Please refer to predata_WSJ_lcx.py A more detailed dataset preparation procedure will be updated soon.

Train and Test

For train:
python train_WSJ0_SDNet.py

For test:
python test_WSJ0_SDNet.py

Please Modify the model path in test_WSJ0_SDNet.py.

Contact

If you have any questions please contact:
Email:[e-mail]

TODO

1. A brief implemention of SDNet
2. pretrained models.
3. separated samples.