Skip to main content

Table 14 Summary of results obtained with i-vector systems and different speaker embeddings in the Tagalog and Cantonese subsets of NIST SRE 2016

From: Introducing phonetic information to speaker embedding for speaker verification

Ā 

Tagalog

Cantonese

Pooled

Ā 

EER(%)

minDCF16

EER(%)

minDCF16

EER(%)

minDCF16

i-vector

21.37

0.8901

10.07

0.6564

15.73

0.7861

DNN/i-vector

22.25

0.9059

11.47

0.6950

16.90

0.8127

x-vector

13.60

0.7877

5.33

0.4429

9.46

0.6365

x-vector-pc (c=0)

13.06

0.7794

4.86

0.4207

8.96

0.6214

x-vector-pv (c=0.2)

12.60

0.7593

4.55

0.4127

8.58

0.6045

x-vector-mt (1-layer sharing)

12.82

0.7629

4.36

0.3878

8.59

0.5941

sc-vector (1-layer sharing)

12.92

0.7675

4.00

0.3835

8.44

0.5951

c-vector (c=0.2 + 1-layer sharing)

12.00

0.7451

4.04

0.3629

8.04

0.5692

  1. The pooled results are also demonstrated