LikeSNU-서울대 중앙도서관 빅데이터 지식 플랫폼

Optimizing a-DCF for Spoofing-Robust Speaker Verification

Author

Kurnaz, O.; Mishra, J.; Kinnunen, T.H.; Hanilci, C.

Journal Title

IEEE Signal Processing Letters

Publication Year

2025

Topics

#Speech recognition
#Computer Science

Summary

This study proposes a method to directly optimize the architecture-agnostic detection cost function (a-DCF) to enhance the robustness of automatic speaker verification (ASV) systems against spoofing attacks, aiming for a balance between user convenience and spoofing prevention. By combining a-DCF and binary cross-entropy with a novel threshold optimization technique, a performance improvement of up to 43% was achieved on the ASVspoof2019 dataset.

Journal Influence

[IEEE Signal Processing Letters]