search

sp-uhh / sgmse

Score-based Generative Models (Diffusion Models) for Speech Enhancement and Dereverberation
MIT License
491 stars 74 forks source link

About the checkpoint #31

Closed yangdongchao closed 1 year ago

yangdongchao commented 1 year ago

Hi, I try to use your checkpoint to inference. To my surprise is that I find the two checkpint (one trained on vb dataset, one trained on wjs dataset) will get the same inference results. The following is my commond:

python enhancement.py --test_dir "SE_data/noisy_testset_wav" \
       --enhanced_dir "test/other_baseline/sgmse/in_domain2" \
       --ckpt  "sgmse/train_vb_29nqe0uh_epoch=115.ckpt"
python enhancement.py --test_dir "SE_data/noisy_testset_wav" \
       --enhanced_dir "test/other_baseline/sgmse/in_domain2" \
       --ckpt  "sgmse/train_wsj0_2cta4cov_epoch=159.ckpt"