Table 4 Speech separation performance as a function of gender pairing and spatial distance between talkers. Kruskal–Wallis H tests with False Discovery Rate (FDR) correction.
Input configuration | Gender pairing | Small distance | Large distance | Improvement | \(\chi ^2\)(1) | Significance |
|---|---|---|---|---|---|---|
SI-SDRi (dB) | SI-SDRi (dB) | (%) | ||||
Two-channel, bilateral | F–F | 5.74 | 8.08 | 40.8 | 64.69 | 1.5e-15*** |
M–M | 6.82 | 8.62 | 26.4 | 89.78 | 1.2e-20*** | |
F–M | 7.73 | 8.91 | 15.4 | 79.27 | 1.4e-18*** | |
Two-channel, bilateral, IPD | F–F | 7.14 | 9.16 | 28.3 | 59.88 | 1.5e-14** |
M–M | 8.24 | 9.73 | 18.1 | 56.31 | 8.5e-14*** | |
F–M | 9.03 | 10.03 | 11.0 | 54.24 | 2.2e-13*** | |
Two-channel, bilateral, ILD | F–F | 5.97 | 8.37 | 40.2 | 65.97 | 9.8e-16*** |
M–M | 6.89 | 8.69 | 26.1 | 87.01 | 3.2e-20*** | |
F–M | 7.66 | 8.94 | 16.7 | 89.46 | 1.1e-20*** | |
Two-channel, bilateral, IPD, ILD | F–F | 6.64 | 9.23 | 39.0 | 72.29 | 4.8e-17*** |
M–M | 7.59 | 9.38 | 23.6 | 79.32 | 1.6e-18*** | |
F–M | 8.42 | 9.70 | 15.3 | 89.48 | 1.9e-20*** |
