From: A simulation study on optimal scores for speaker recognition
 |  | x-vector | i-vector | ||
---|---|---|---|---|---|
 |  | True data | Synthesis data | True data | Synthesis data |
Within class | PC1 dir. STD | 0.095 | 0.045 | 0.060 | 0.045 |
 | PC2 dir. STD | 0.077 | 0.045 | 0.063 | 0.055 |
 | Avg PC dir. STD | 0.084 | 0.045 | 0.055 | 0.055 |
 | PC1 shape STD | 11.2 | 1.80 | 11.5 | 1.94 |
 | PC2 shape STD | 6.83 | 1.82 | 7.95 | 1.95 |
 | Avg PC shape STD | 5.53 | 1.85 | 5.48 | 1.98 |
 | PC Kurtosis | 19.15 | 0.150 | 1.940 | 0.152 |
 | PC Skewness | 0.837 | 0.040 | 0.370 | 0.041 |
Between class | PC Kurtosis | 0.390 | 0.010 | 0.358 | 0.002 |
 | PC Skewness | 0.065 | 0.001 | 0.043 | 0.001 |
 | EER% (STD) | 2.77 (0.1) | 0.0 (0.0) | 3.22 (0.2) | 0.003 (0.008) |
 | IDR% (STD) | 85.51(0.7) | 100.0 (0.0) | 75.08 (0.5) | 100.0(0.0) |