sp-uhh / sgmse

Score-based Generative Models (Diffusion Models) for Speech Enhancement and Dereverberation
MIT License
454 stars 69 forks source link

How to use dereverbration checkpoint for inference? #9

Closed the6thsense closed 1 year ago

the6thsense commented 1 year ago

Please suggest a bash command to run the dereverbration checkpoint.

julius-richter commented 1 year ago

That would be

python enhancement.py --test_dir <your_test_dir> --enhanced_dir <your_enhanced_dir> --ckpt <path_to_model_checkpoint> --N 50 --snr 0.33

the6thsense commented 1 year ago

Thanks @julius-richter, Is it that by just changing the checkpoint file from enhancement one to dereverbration one the model will itself identify which model to load?

cobalamin commented 1 year ago

Hi, the only difference between the two models is really the training data (and therefore the weights), as well as some minor changes in the parameterization of the stochastic process. When using ScoreModel.load_from_checkpoint(...) as enhancement.py does, it will automatically load all of this correctly and be usable for the respective task each model was trained for.

the6thsense commented 1 year ago

That sounds interesting. Thanks @cobalamin