Skip to main content
. Author manuscript; available in PMC: 2016 Dec 1.
Published in final edited form as: IEEE/ACM Trans Audio Speech Lang Process. 2016 Mar 1;24(5):967–977. doi: 10.1109/TASLP.2016.2536478

TABLE I.

STOI (in Percent) Comparison Between Speech Separation Methods With Single-SNR Speaker-Pair Dependent Training on SSC Corpus. The Results Are Averaged Over 8 Speaker Pairs

−12 dB −9 dB −6 dB −3 dB O dB 3 dB 6 dB
Noisy 46.4 52.2 58.8 65.8 72.7 79.2 84.8
DNN+Map 71.4 76.7 81.0 84.9 88.3 91.2 93.5
DNN+IRM 72.1 76.5 80.5 84.5 87.8 90.9 93.2
MCA+IRM 73.9 78.3 82.2 86.0 89.1 91.7 93.9
MCS+IRM 73.7 78.5 82.4 86.3 89.4 92.1 94.2
DNN+SA 76.9 81.0 84.7 88.0 90.8 93.3 95.3
MCA+SA 77.8 82.2 85.8 88.9 91.6 93.8 95.6
MCS+SA 78.4 82.2 86.1 88.9 91.8 93.9 95.7