A dual path convolutional neural network model for high accuracy Blind Source Separation (BSS).
- Downloading MUSDB18 dataset (we recommend you to download .wav version or transform the mp4 to .wav format).
- Converting the wav files to h5 files:
python ./dataset/make_dataset, to speed up training. - Training a Y-Net model:
python train.py. - Modifying the structure of the Y-Net model at
./configs/defaults.py. - Modifying the function
train_cfg()at./train.pyto change the hyperparameters of the Y-Net model. - Defying the target output sources at
./train.py INSTRUMENTS. - Modifying the path to validate the Y-Net model:
python validate.py.
If you like our repository, please cite our papers.
@INPROCEEDINGS{Wu2012:Y,
AUTHOR={Huanzhuo Wu and Jia He and M{\'a}t{\'e {T{\"o}m{\"o}sk{\"o}zi} and Frank H.P. Fitzek},
TITLE="{Y-Net:} A Dual Path Model for High Accuracy Blind Source Separation",
BOOKTITLE="2020 IEEE Globecom Workshops (GC Wkshps): IEEE GLOBECOM 2020 Workshop on Future of Wireless Access for Industrial IoT (FutureIIoT) (GC 2020 Workshop - FIIoT)",
ADDRESS="Taipei, Taiwan",
DAYS=6,
MONTH=dec,
YEAR=2020
}
We are researchers at the Deutsche Telekom Chair of Communication Networks (ComNets) at TU Dresden, Germany. Our focus is on in-network computing.
- Huanzhuo Wu - huanzhuo.wu@tu-dresden.de
- Jia He - jia.he@mailbox.tu-dresden.de
This project is licensed under the MIT license.