Giters

etzinis / sudo_rm_rf

Code for SuDoRm-Rf networks for efficient audio source separation. SuDoRm-Rf stands for SUccessive DOwnsampling and Resampling of Multi-Resolution Features which enables a more efficient way of separating sources from mixtures.

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Few questions

q121q opened this issue · comments

commented

Hi,

Thanks for publishing your code, have a few questions I hope are of interest to the greater community:
a) Do you have results for libriMix? Looks like there's code support for that, IIRC the paper doesn't mention any results on it.
b) How long should it take to train? Would a single 2080 be enough to train on libriMix in one day?
c) Can you provide any pretrained checkpoints?

Thanks

commented

Thanks for reaching out.
a) I had some initial experiments but did not test fully since I just did this experiment to see what am I getting. THe results were similar to other architectures.
b) It depends on the level of convergence and your model. You should be able to get reasonable numbers after 1 day of training.
c) Working towards that, thanks for bringing that up!

commented

Please check my latest commit containign anechoic and noisy/reverberant mixture checkpoints.