Skip to main content

Advertisement

Table 1 Male speaker verification results of GMM-UBM method in terms of percent EER (minDCF) for the proposed algorithm, Drugman’s VAD method [27], and Rangachari’s noise tracking method [21]. The last columns show the relative percent EER reduction rates compared to Drugman’s VAD and Rangachari’s method, respectively

From: A robust polynomial regression-based voice activity detector for speaker verification

Noise type SNR level (dB) Proposed algorithm Drugman’s VAD Rangachari’s noise tracking EER reduction compared to Drugman’s EER reduction compared to Rangachari’s
Lynx − 10 34.25 (0.64) 46.10 (0.85) 47.4 (0.87) 25.70 27.74
− 5 25.30 (0.47) 32.18 (0.60) 39.22 (0.72) 21.38 35.49
0 15.29 (0.28) 14.60 (0.27) 22.47 (0.42) − 4.72 31.95
5 8.41 (0.15) 8.41 (0.15) 13.45 (0.24) 0 37.47
10 5.42 (0.10) 6.50 (0.12) 9.93 (0.18) 16.61 45.41
F16 − 10 41.28 (0.78) 48.31 (0.88) 48.16 (0.89) 14.55 14.28
− 5 31.88 (0.60) 41.82 (0.80) 45.18 (0.84) 23.77 29.43
0 20.87 (0.39) 24.38 (0.46) 33.4 (0.60) 14.40 37.51
5 11.85 (0.22) 11.54 (0.21) 18.19 (0.34) − 2.68 34.85
10 6.95 (0.13) 7.8 (0.14) 12.46 (0.23) 10.89 44.22
Car − 10 5.96 (0.10) 6.27 (0.11) 8.94 (0.16) 4.94 33.33
− 5 4.74 (0.08) 5.88 (0.10) 8.35 (0.15) 19.38 43.23
0 4.35 (0.08) 5.50 (0.10) 8.18 (0.15) 20.91 46.82
5 4.05 (0.07) 5.27 (0.09) 7.95 (0.14) 23.15 49.05
10 4.05 (0.07) 5.12 (0.09) 7.95 (0.14) 20.90 49.05
Babble − 10 36.85 (0.69) 48.08 (0.87) 47.85 (0.88) 23.35 22.98
− 5 26.83 (0.50) 38.45 (0.72) 43.94 (0.87) 30.22 42.84
0 17.50 (0.33) 19.49 (0.36) 28.28 (0.51) 10.21 38.11
5 10.01 (0.18) 10.16 (0.18) 14.52 (0.27) 1.47 31.06
10 6.72 (0.12) 7.26 (0.13) 10.93 (0.20) 7.44 38.51
Stitel − 10 42.66 (0.79) 47.17 (0.86) 45.18 (0.84) 9.56 5.57
− 5 33.71 (0.62) 37.23 (0.69) 37.15 (0.69) 9.45 9.26
0 19.95 (0.37) 19.26 (0.36) 20.41 (0.38) − 3.58 2.25
5 9.40 (0.17) 9.71 (0.18) 11.62 (0.21) 3.19 19.10
10 5.96 (0.11) 6.95 (0.12) 9.32 (0.17) 14.24 36.05