From: An investigation of supervector regression for forensic voice comparison on small data
 |  | 60 female speaker database | 90 male speaker database | ||
---|---|---|---|---|---|
Validity ( C llr mean) | Reliability (95% CI) | Validity ( C llr mean) | Reliability (95% CI) | ||
GMM-UBM | Gradient projection, s x(1) | 48.5% | 21.5% | 63.5% | 33.8% |
LS, s x(1) | 49.7% | 19.1% | 63.6% | 31.3% | |
Gradient projection, \( {s}_{\ell_1\mathrm{norm}} \) | 75.0% | 32.3% | 50.5% | 41.5% | |
LS, \( {s}_{\ell_1\mathrm{norm}} \) | 75.1% | 32.7% | 50.8% | 41.5% | |
s SVM | Gradient projection, s x(1) | 48.9% | 14.1% | 11.6% | 9.1% |
LS, s x(1) | 50.1% | 11.5% | 11.8% | 5.7% | |
Gradient projection, \( {s}_{\ell_1\mathrm{norm}} \) | 75.1% | 26.0% | −19.9% | 19.7% | |
LS, \( {s}_{\ell_1\mathrm{norm}} \) | 75.3% | 26.4% | −19.1% | 19.7% |