variance-preserving-based interpolation diffusion models for speech enhancement, in which we apply the diffusion model to the speech enhancement (denoising) task. Diffusion Models for Speech Enhancement.
Install requirements in requirements.txt via
pip install -r requirements.txt
python train.py --base_dir <your vbd dataset dir>
--gpus 4
--no_wandb
--sde vpsde
--eta 1.5
--beta-max 2
--N 25
--t_eps 4e-2
--logdir <your log dir>
python enhancement.py --test_dir exp_dir
--corrector_step 0
--N 25
--enhanced_dir /outputs
--ckpt <your checkpoint best.ckpt>