Skip to main content

Table 2 Improvements over the GMM-UBM and SVM baseline systems

From: An investigation of supervector regression for forensic voice comparison on small data

  

60 female speaker database (degraded conditions)

60 female speaker database (mismatched conditions)

  

Validity ( C llr mean)

Reliability (95% CI)

Validity ( C llr mean)

Reliability (95% CI)

GMM-UBM

Gradient projection, \( {s}_{\ell_1\mathrm{norm}} \)

35.1%

0.3%

23.1%

7.0%

LS, \( {s}_{\ell_1\mathrm{norm}} \)

37.9%

1.1%

24.0%

6.8%

s SVM

Gradient projection, \( {s}_{\ell_1\mathrm{norm}} \)

38.9%

11.4%

10.6%

2.8%

LS, \( {s}_{\ell_1\mathrm{norm}} \)

41.5%

12.1%

11.6%

2.5%

  1. In terms of percentage for the pooled results of validity (C llr mean) and reliability (95% CI) across the six permutations for the two systems: gradient projection, s â„“1norm; and LS, s â„“1norm evaluated on 60 female speaker database under degraded and mismatched conditions.