Enhancing Speaker Verification with w2v-BERT 2.0 and Knowledge Distillation guided Structured Pruning

Benchmark Model Rank Results
speaker-verification-on-voxceleb1-e-cleanw2v-BERT 2.0 LoRA Adapter MFA + LMFT + calibration (extra data)–EER: 0.27minDCF (P_target=0.01): 0.028
speaker-verification-on-voxceleb1-h-cleanw2v-BERT 2.0 LoRA Adapter MFA + LMFT + calibration (extra data)–EER: 0.55minDCF (P_target=0.01): 0.051
speaker-verification-on-voxceleb1-o-cleanw2v-BERT 2.0 LoRA Adapter MFA + LMFT + calibration (extra data)–EER: 0.12minDCF (P_target=0.01): 0.025