The official code for Learnable spectral dimension compression mapping for full-band speech enhancement (2023), https://asa.scitation.org/doi/10.1121/10.0017327. The code for fullsubnet-plus is taken from https://github.com/hit-thusz-RookieCJ/FullSubNet-plus.
- pytorch 1.7.1
- librosa
- soundfile
- pesq
- pystoi
git clone https://github.com/Qinwen-Hu/SDCM.git
Download the VoiceBank-Demand dataset from https://datashare.ed.ac.uk/handle/10283/2791, modify the corresponding yaml document, and run python Dataloader_vctk_demand.py to split the data into clips.
Run python train_fullsubp.py or python train_dptfsnet.py.
eg. Run:
python evaluate.py -m f_scm -c ./ckpt/fullsubp_scm/model.ckpt -N /data/ssd/vctk_demand/noisy_testset_wav -C /data/ssd/vctk_demand/clean_testset_wav -O ./vctk_demand_enhanced -d cpu
@article{sdcm2023,
title={Learnable spectral dimension compression mapping for full-band speech enhancement},
author={Hu, Qinwen and Hou, Zhongshu and Chen, Kai and Lu, Jing},
journal={JASA Express Letters},
volume={3},
number={2},
pages={025204},
year={2023},
publisher={Acoustical Society of America}
}