TY - GEN
T1 - Improving Generalization for AI-Synthesized Voice Detection
AU - Ren, Hainan
AU - Lin, Li
AU - Liu, Chun Hao
AU - Wang, Xin
AU - Hu, Shu
N1 - Publisher Copyright: Copyright © 2025, Association for the Advancement of Artificia Intelligence (www.aaai.org). All rights reserved.
PY - 2025/4/11
Y1 - 2025/4/11
N2 - AI-synthesized voice technology has the potential to create realistic human voices for beneficial applications, but it can also be misused for malicious purposes. While existing AI-synthesized voice detection models excel in intra-domain evaluation, they face challenges in generalizing across different domains, potentially becoming obsolete as new voice generators emerge. Current solutions use diverse data and advanced machine learning techniques (e.g., domain-invariant representation, self-supervised learning), but are limited by predefined vocoders and sensitivity to factors like background noise and speaker identity. In this work, we introduce an innovative disentanglement framework aimed at extracting domain-agnostic artifact features related to vocoders. Utilizing these features, we enhance model learning in a flat loss landscape, enabling escape from suboptimal solutions and improving generalization. Extensive experiments on benchmarks show our approach outperforms state-of-the-art methods, achieving up to 5.12% improvement in the equal error rate metric in intra-domain and 7.59% in cross-domain evaluations.
AB - AI-synthesized voice technology has the potential to create realistic human voices for beneficial applications, but it can also be misused for malicious purposes. While existing AI-synthesized voice detection models excel in intra-domain evaluation, they face challenges in generalizing across different domains, potentially becoming obsolete as new voice generators emerge. Current solutions use diverse data and advanced machine learning techniques (e.g., domain-invariant representation, self-supervised learning), but are limited by predefined vocoders and sensitivity to factors like background noise and speaker identity. In this work, we introduce an innovative disentanglement framework aimed at extracting domain-agnostic artifact features related to vocoders. Utilizing these features, we enhance model learning in a flat loss landscape, enabling escape from suboptimal solutions and improving generalization. Extensive experiments on benchmarks show our approach outperforms state-of-the-art methods, achieving up to 5.12% improvement in the equal error rate metric in intra-domain and 7.59% in cross-domain evaluations.
UR - https://www.scopus.com/pages/publications/105004322552
U2 - 10.1609/aaai.v39i19.34221
DO - 10.1609/aaai.v39i19.34221
M3 - Conference contribution
T3 - Proceedings of the AAAI Conference on Artificial Intelligence
SP - 20165
EP - 20173
BT - Special Track on AI Alignment
A2 - Walsh, Toby
A2 - Shah, Julie
A2 - Kolter, Zico
PB - Association for the Advancement of Artificial Intelligence
T2 - 39th Annual AAAI Conference on Artificial Intelligence, AAAI 2025
Y2 - 25 February 2025 through 4 March 2025
ER -