lightning_dcunet

Speech enchancement without labels. Coded with pytorch lightning.

Setup

You need to install following libraries

numpy
pandas
matplotlib
seaborn
scipy
pesq
pytorch-lightning
lightning
tensorboard
torchinfo
thefuck
librosa
rich
black
pydub
python-dotenv

Start up

args

arg	option	default	required
--mode	train , predict	None	yes
--dataset	white , urban0 , urban1 , urban2	None	yes
--loss	nct , nb2nb	None	yes
--model	dcunet , dcunet-rtstm	None	yes
--devices	0 ~ 8	0	no
--checkpoint	your checkpoint path	None	no

exapmple

python3 trainlpy --mode train --dataset white --loss nb2nb --model dcunet --devices 0 1 2 3 4 5 6 7