"EFFECTIVE LOW-COST TIME-DOMAIN AUDIO SEPARATION USING GLOBALLY ATTENTIVE LOCALLY RECURRENT NETWORKS" paper link:https://arxiv.org/pdf/2101.05014v1.pdf
tasnet is the basic of recent speech separation! convtasnet is also a good job! Today GALR is similar to DPRNN,DPTnet,perhaps all of them are good at long term modeling but i recommend the sudoRMRF!
i am not the author of this paper,so there are some defects in GALR.py first,the poistion encoding second,the dimension is little bit different from paper
if there are some errors,please contact me ! thanks the former contributors had been mentioned in GALR.py