From: Introducing phonetic information to speaker embedding for speaker verification
Ā | Core-extended | 10āsā10ās | ||||
---|---|---|---|---|---|---|
Ā | EER(%) | minDCF08 | minDCF10 | EER(%) | minDCF08 | minDCF10 |
i-vector | 2.33 | 0.0127 | 0.4132 | 10.29 | 0.0521 | 0.9695 |
DNN/i-vector | 0.89 | 0.0047 | 0.1969 | 7.25 | 0.0334 | 0.9160 |
x-vector | 1.96 | 0.0109 | 0.3900 | 7.62 | 0.0428 | 0.8321 |
x-vector-pa (c=0) | 1.68 | 0.0095 | 0.3585 | 6.86 | 0.0406 | 0.8053 |
x-vector-pa (c=0.1) | 1.44 | 0.0084 | 0.2979 | 6.45 | 0.0385 | 0.7366 |
x-vector-mt (3-layer sharing) | 1.52 | 0.0073 | 0.2752 | 6.10 | 0.0369 | 0.8808 |
sc-vector (3-layer sharing) | 1.24 | 0.0071 | 0.3035 | 6.80 | 0.0351 | 0.7854 |
c-vector (c=0.1 + 3-layer sharing) | 1.21 | 0.0065 | 0.2449 | 6.40 | 0.0334 | 0.5878 |