Open-set speaker identification on ESD (test)
0.61EERSpeakerRPL V2 (proposed)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| SpeakerRPL V2 (proposed)Target speaker augmentation=true, Unknown speaker augmentation=true, Adaptive anchor=true, Model Fusion=true2026.04 | 0.61 | 0.04 | 96.63 | 98.63 | |
| SpeakerRPL V2 w/ naive fusionTarget speaker augmentation=true, Unknown speaker augmentation=true, Adaptive anchor=true, Model Fusion=true2026.04 | 0.63 | 0.04 | 96.6 | 98.61 | |
| SpeakerRPL V2 w/o fusionTarget speaker augmentation=true, Unknown speaker augmentation=true, Adaptive anchor=true, Model Fusion=false2026.04 | 0.94 | 0.06 | 94.93 | 97.71 | |
| AM-SoftmaxTarget speaker augmentation=true, Unknown speaker augmentation=true, Adaptive anchor=true, Model Fusion=false2026.04 | 1.04 | 0.06 | 94.23 | 97.15 | |
| AAM-SoftmaxTarget speaker augmentation=true, Unknown speaker augmentation=true, Adaptive anchor=true, Model Fusion=false2026.04 | 1.04 | 0.06 | 94 | 97.16 | |
| SpeakerRPL V1Target speaker augmentation=true, Unknown speaker augmentation=true, Adaptive anchor=false, Model Fusion=false2026.04 | 1.31 | 0.08 | 94.86 | 97.75 | |
| SoftmaxTarget speaker augmentation=true, Unknown speaker augmentation=true, Adaptive anchor=true, Model Fusion=false2026.04 | 1.47 | 0.08 | 92.26 | 96.02 | |
| PrototypeTarget speaker augmentation=true, Unknown speaker augmentation=true, Adaptive anchor=true, Model Fusion=false2026.04 | 1.67 | 0.09 | 91.21 | 95.2 | |
| Direct Enrollment +Target augTarget speaker augmentation=true, Unknown speaker augmentation=false, Adaptive anchor=false, Model Fusion=false2026.04 | 4.28 | 0.22 | 96.27 | 98.71 | |
| Direct EnrollmentTarget speaker augmentation=false, Unknown speaker augmentation=false, Adaptive anchor=false, Model Fusion=false2026.04 | 4.33 | 0.22 | 96.17 | 98.63 | |
| SpeakerRPL V1 w/o Unknown speakersTarget speaker augmentation=true, Unknown speaker augmentation=false, Adaptive anchor=false, Model Fusion=false2026.04 | 5.28 | 0.31 | 93.12 | 98.79 | |
| SpeakerRPL V1 w/o Target augTarget speaker augmentation=false, Unknown speaker augmentation=false, Adaptive anchor=false, Model Fusion=false2026.04 | 5.74 | 0.32 | 92.19 | 98.23 |