From: Learning-based robust speaker counting and separation with the aid of spatial coherence
MAC
sG1
G1
G2
G3
1
0.989
0.986
0.978
0.975
0.993