Add Readme and pretrained model to fasNet recipe

asteroid-team / asteroid

The PyTorch-based audio source separation toolkit for researchers

https://asteroid-team.github.io/

MIT License

2.22k stars 422 forks source link

Add Readme and pretrained model to fasNet recipe #561

Closed JorisCos closed 2 years ago

JorisCos commented 2 years ago

I have trained a model with the FasNet recipe and thought it would be good to add results and pretrained model. @popcornell I have run the experiment with the default recipe but I am a bit confused on how to compare it to the paper.

If I'm not mistaken we are reproducing the last row of table 1 in this configuration. However, I am confused about about the overlap part is it the mean of everything ?

The result gives an improvement of 9.9 dB SI-SNR

popcornell commented 2 years ago

The average in the paper should be the result on all the test set. The other results are obtained by selecting a subset of the total mixtures with every subset composed of mixtures with < X overlap ratio.