sp-uhh / sgmse

Score-based Generative Models (Diffusion Models) for Speech Enhancement and Dereverberation
MIT License
520 stars 76 forks source link

what is the test dir? #54

Closed FrankCasbert closed 2 weeks ago

FrankCasbert commented 2 months ago

python calc_metrics.py --test_dir <your_test_dir> --enhanced_dir <your_enhanced_dir>

The calc_metrics.py needs three directories: clean dir, noisy dir and enhanced dir. However, the guideline say --test_dir <your_test_dir>

So how to enter my file, I'm confused. Should I enter 'python calc_metrics.py --clean_dir test/clean --noisy_dir test/noisy_dir ...'

thanks

grepLord commented 1 month ago

yes I had the same query. I also tried the modified command. It worked.

I have some doubt regarding the enhanced files. Why is their two folders inside enhanced dir. one is clean and other is noisy. Is it not supposed to generate just an enhanced output from the given clean and noisy input files.

Kindly respond at the earliest.

Thanks.

julius-richter commented 3 weeks ago

What do you mean by "Why is there two folders inside enhanced dir"?

grepLord commented 3 weeks ago

Thanks for the reply.

What I mean is that, when noisy files and clean files are given as tests cases to the model, it is supposed to generate enhanced speech files. Now it does generates them but along with it, it also has noisy dir in the enhancement dir. So I have clean dir and noisy dir inside the enhancement dir. Is this correct?

julius-richter commented 3 weeks ago

The clean files are not given to the model for enhancement (see enhancement.py https://github.com/sp-uhh/sgmse/blob/main/enhancement.py#L38).

The clean files are only used to calculate the intrusive speech metrics (see calc_metrics.py https://github.com/sp-uhh/sgmse/blob/main/calc_metrics.py).

Enhancement dir contains the enhanced files only.

Does that make sense?