Table 3.
EERM, EERF, and HEER obtained on development set (no score normalization), comparing classical parameters with extra parameters and extended biometric parameters for the configurations providing the most successful results [RR → Relative Reduction/[threshold]/(p-value)].
| Parameters | Genre | GSE + VTE setup | Extra parameters | EERM [θM] (p-value) | EERM RR | EERF [θF] (p-value) | EERF RR | HEER [RR] |
|---|---|---|---|---|---|---|---|---|
| Gender-independent configuration (GIC MFCCs + Δ) | M/F | – | – | 2.534% [−0.178] | – | 2.170% [–0.169] | – | 2.352% [–] |
| Gender-dependent configuration (GDC MFCCs + Δ) | M | – | – | 2.390% [−0.001] (5.57 × 10−1) | 5.68% | 2.193% [6.76%] | ||
| F | – | – | 1.996% [−0.166] (4.07 × 10−1) | 8.02% | ||||
| Gender-dependent configuration (GDC MFCCs + Δ + Extra) | M | – | E + ΔE | 2.163% [−0.035] (2.09 × 10−1) | 14.64% | 1.991% [15.37%] | ||
| F | – | E + ΔE + F0 + F3 | 1.818% [−0.113] (2.12 × 10−1) | 16.23% | ||||
| Gender-dependent configuration (GDC MFCCs + Δ + Extra + GSE) | M |
Source-tract sep. Alg Prediction order: 10 Forgetting factor: 0.995 GSE 13-Channel Filter bank 8 MFCC |
E + ΔE | 1.504% [−0.131] (5.41 × 10−4) | 40.65% | 1.477% [37.19%] | ||
| F |
Source-tract sep. Alg Prediction order:16 Forgetting factor: 0.995 GSE 12-Channel Filter bank 4 MFCC |
E + ΔE + F0 + F3 | 1.451% [−0.145] (1.67 × 10−3) | 33.15% | ||||
The results highlighted in green, are the ones achieving higher recognition rates.