최신 ASV에서의 적대적 공격과 방어: a-DCF 기반 성능 평가

Vol. 35, No. 2, pp. 277-286, 4월. 2025
10.13089/JKIISC.2025.35.2.277, Full Text:
Keywords: Automatic speaker verification, adversarial attacks, Biometric Security, a-DCF
Abstract

This study explores evaluation methods to enhance the security of ASV (Automatic Speaker Verification) systems in adversarial attack environments, where existing research is limited. To address the limitations of the conventional DCF (Detection Cost Function), the a-DCF (agnostic-Detection Cost Function) from the SASV (Spoofing-Aware Speaker Verification) field was adopted. Representative adversarial attack methods were applied to the latest ASV models, NeXt-TDNN and ECAPA-TDNN, to analyze attack success rates and adversarial sample quality. Additionally, the MEH-FEST (Minimum Energy in High Frequencies for Short Time) detection method, based on high-frequency energy, was implemented to examine defense effectiveness. This study aims to validate the utility of a-DCF and contribute to enhancing the reliability and security of ASV systems.

Statistics
Show / Hide Statistics

Statistics (Cumulative Counts from December 1st, 2017)
Multiple requests among the same browser session are counted as one view.
If you mouse over a chart, the values of data points will be shown.


Cite this article
[IEEE Style]
이요원, 정수환, 홍기훈, "Adversarial Attacks and Defense in ASV:a-DCF-Based Performance Evaluation," Journal of The Korea Institute of Information Security and Cryptology, vol. 35, no. 2, pp. 277-286, 2025. DOI: 10.13089/JKIISC.2025.35.2.277.

[ACM Style]
이요원, 정수환, and 홍기훈. 2025. Adversarial Attacks and Defense in ASV:a-DCF-Based Performance Evaluation. Journal of The Korea Institute of Information Security and Cryptology, 35, 2, (2025), 277-286. DOI: 10.13089/JKIISC.2025.35.2.277.