unilight / sheet

Speech Human Evaluation Estimation Toolkit (SHEET)
MIT License
43 stars 5 forks source link

use cuda for prediction #2

Closed giangtranducts closed 3 weeks ago

giangtranducts commented 3 weeks ago

May you guide me to use predict in cuda for better performance? Thanks a lot (https://github.com/unilight/sheet?tab=readme-ov-file#i-just-want-to-use-your-trained-mos-predictor)

sh-lee-prml commented 3 weeks ago
    ssl_mos_predictor = torch.hub.load("unilight/sheet:v0.1.0", "default", trust_repo=True, force_reload=True)
    ssl_mos_predictor.model.cuda() # Add this line

    result= ssl_mos_predictor.predict(wav=torch.rand(16000).cuda())

I modified the code for CUDA. Try to use this :)

giangtranducts commented 3 weeks ago

Thank a lot @sh-lee-prml

unilight commented 3 weeks ago

@sh-lee-prml Thanks a lot! I will modify the readme with your suggestion :)

sh-lee-prml commented 3 weeks ago

@unilight

Thanks for sharing nice speech quality estimator. I'm currently using this with UTMOS together :)

It would be a good reference for model evaluation.

Thanks