EMBNet: Multi‐scale feature learning with efficient channel attention for deepfake speech detection
Journal of Forensic Sciences · 2026-11 · 71 (6) · 2772-2786
DOI: 10.1111/1556-4029.70432 ↗Kurz gefasst
Fachartikel in der Zeitschrift Journal of Forensic Sciences, online erschienen am 10.08.2026. Verfasst von Haitao Yang und 6 weiteren Autor:innen. Beteiligte Einrichtung: Department of Criminal Investigation, Hunan Police Academy, Changsha, Hunan, China. Hauptthemen laut MEDLINE: Speech Acoustics, Deep Learning. Die Arbeit nennt 32 Literaturangaben. Gefördert von: The Science and Technology Program of Hunan Province, Scientific Research Project of the Hunan Provincial Department of Education, Science and Technology Program of Hunan Province.
Abstract (Originalsprache)
With the rapid advancement of generative artificial intelligence, deepfake speech has emerged as a significant threat to digital audio authenticity, posing challenges for forensic analysis and legal applications. In this study, we propose EMBNet, a task‐oriented deepfake speech detection framework that integrates efficient channel attention (ECA) with a multi‐scale bottleneck to enhance the representation of subtle acoustic anomalies. The ECA module adaptively emphasizes critical feature channels, while the multi‐scale bottleneck captures local and hierarchical spoofing traces across multiple time–frequency resolutions. The proposed framework effectively balances fine‐grained local detail modeling with global hierarchical representation, improving the detection of weakly manifested spoofing patterns. Extensive experiments on the ASVspoof 2019 Logical Access dataset demonstrate that EMBNet significantly outperforms existing baseline and state‐of‐the‐art methods, achieving an EER of 2.67%, an AUC of 97.32%, and an F1‐score of 97.28%. Ablation studies further confirm the complementary contributions of the ECA and multi‐scale modules to overall performance. The proposed approach demonstrates promising performance for forensic audio analysis and may support forensic audio authenticity assessment by improving the discrimination between genuine and manipulated speech.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Department of Criminal Investigation, Hunan Police Academy, Changsha, Hunan, China.
- Hainan Open University, Haikou, Hainan, China.
- Zhengzhou Police University, Zhengzhou, Henan, China.
- School of Public Security Information Technology and Intelligence, Criminal Investigation Police University of China, Shenyang, Liaoning, China.
Förderung
- The Science and Technology Program of Hunan Province
- Scientific Research Project of the Hunan Provincial Department of Education
- Science and Technology Program of Hunan Province
- Chongqing Natural Science Foundation Joint Innovation Development Fund (Science and Technology Commission) Class A Project.
- Scientific and technological project in Henan Province
Zitieren
APA
Yang, H., Li, F., Cai, X., Huang, J., Tang, Y., Xiong, Y., & Wang, H. (2026). EMBNet: Multi‐scale feature learning with efficient channel attention for deepfake speech detection. Journal of Forensic Sciences, 71(6), 2772-2786. https://doi.org/10.1111/1556-4029.70432
Vancouver
Yang H, Li F, Cai X, Huang J, Tang Y, Xiong Y, et al. EMBNet: Multi‐scale feature learning with efficient channel attention for deepfake speech detection. Journal of Forensic Sciences. 2026;71(6):2772-2786. doi:10.1111/1556-4029.70432
BibTeX
@article{Yang2026,
author = {Yang, Haitao and Li, Fen and Cai, Xin and Huang, Juanjuan and Tang, Yifan and Xiong, Yingzhuo and Wang, Huapeng},
title = {EMBNet: Multi‐scale feature learning with efficient channel attention for deepfake speech detection},
journal = {Journal of Forensic Sciences},
year = {2026},
volume = {71},
number = {6},
pages = {2772--2786},
doi = {10.1111/1556-4029.70432}
}
EUVIMED