Speaker Verification on VoxCeleb2 (Vox-H test)
2.239EERE.-H. AM-Softmax
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| E.-H. AM-SoftmaxMargin=true, Scale=true2026.01 | 2.239 | 0.139 | |
| Real AM-SoftmaxMargin=true, Scale=true2026.01 | 2.258 | 0.137 | |
| HAM-SoftmaxMargin=true, Scale=true2026.01 | 2.268 | 0.143 | |
| AAM-SoftmaxMargin=true, Scale=true2026.01 | 2.662 | 0.163 | |
| OR-Gatenoise_ratio (η)=5%2022.11 | 2.71 | — | |
| AM-SoftmaxMargin=true, Scale=true2026.01 | 2.732 | 0.168 | |
| OR-Gatenoise_ratio (η)=10%2022.11 | 2.77 | — | |
| OR-Gatenoise_ratio (η)=30%2022.11 | 2.87 | — | |
| H-SoftmaxMargin=false, Scale=true2026.01 | 2.889 | 0.18 | |
| Baselinenoise_ratio (η)=0%2022.11 | 2.89 | — | |
| OR-Gatenoise_ratio (η)=0%2022.11 | 2.93 | — | |
| OR-Gatenoise_ratio (η)=20%2022.11 | 2.96 | — | |
| LNCL+Sub-AM + Cosinenoise_ratio (η)=0%, scoring_method=Cosine2022.11 | 3.02 | — | |
| LNCL+Sub-AM + NL-PLDAnoise_ratio (η)=0%, scoring_method=NL-PLDA2022.11 | 3.02 | — | |
| LNCL+Sub-AM + PLDAnoise_ratio (η)=0%, scoring_method=PLDA2022.11 | 3.03 | — | |
| LNCL+Sub-AM + Cosinenoise_ratio (η)=5%, scoring_method=Cosine2022.11 | 3.05 | — | |
| LNCL+Sub-AM + Cosinenoise_ratio (η)=10%, scoring_method=Cosine2022.11 | 3.06 | — | |
| LNCL+Sub-AM + NL-PLDAnoise_ratio (η)=5%, scoring_method=NL-PLDA2022.11 | 3.07 | — | |
| LNCL+Sub-AM + PLDAnoise_ratio (η)=5%, scoring_method=PLDA2022.11 | 3.1 | — | |
| LNCL+Sub-AM + NL-PLDAnoise_ratio (η)=10%, scoring_method=NL-PLDA2022.11 | 3.1 | — | |
| Baselinenoise_ratio (η)=5%2022.11 | 3.11 | — | |
| OR-Gatenoise_ratio (η)=50%2022.11 | 3.11 | — | |
| LNCL+Sub-AM + PLDAnoise_ratio (η)=10%, scoring_method=PLDA2022.11 | 3.15 | — | |
| LNCL+Sub-AM + NL-PLDAnoise_ratio (η)=20%, scoring_method=NL-PLDA2022.11 | 3.2 | — | |
| Baselinenoise_ratio (η)=10%2022.11 | 3.25 | — | |
| LNCL+Sub-AM + Cosinenoise_ratio (η)=20%, scoring_method=Cosine2022.11 | 3.26 | — | |
| LNCL+Sub-AM + Cosinenoise_ratio (η)=30%, scoring_method=Cosine2022.11 | 3.26 | — | |
| LNCL+Sub-AM + NL-PLDAnoise_ratio (η)=30%, scoring_method=NL-PLDA2022.11 | 3.33 | — | |
| LNCL+Sub-AM + PLDAnoise_ratio (η)=20%, scoring_method=PLDA2022.11 | 3.57 | — | |
| LNCL+Sub-AM + NL-PLDAnoise_ratio (η)=50%, scoring_method=NL-PLDA2022.11 | 3.58 | — | |
| Baselinenoise_ratio (η)=20%2022.11 | 3.64 | — | |
| LNCL+Sub-AM + PLDAnoise_ratio (η)=30%, scoring_method=PLDA2022.11 | 3.76 | — | |
| LNCL+Sub-AM + Cosinenoise_ratio (η)=50%, scoring_method=Cosine2022.11 | 3.77 | — | |
| SoftmaxMargin=false, Scale=false2026.01 | 4.321 | 0.255 | |
| Baselinenoise_ratio (η)=30%2022.11 | 4.42 | — | |
| LNCL+Sub-AM + PLDAnoise_ratio (η)=50%, scoring_method=PLDA2022.11 | 4.76 | — | |
| SoftmaxMargin=false, Scale=true2026.01 | 6.082 | 0.353 | |
| Baselinenoise_ratio (η)=50%2022.11 | 6.74 | — |