Semantic-Aware Self-Supervised Distillation for Efficient Real-Time Voice Activity Detection Systems
International Journal on Semantic Web and Information Systems · 20.03.2026 · 22 (1) · 1-18
DOI: 10.4018/ijswis.405058 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal on Semantic Web and Information Systems, erschienen am 20.03.2026. Verfasst von Zhetao Lyu und 4 weiteren Autor:innen. Die Arbeit nennt 48 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 3.0).
Abstract (Originalsprache)
Voice activity detection (VAD) plays a crucial role in speech processing systems, determining the presence of speech in an audio signal. The accuracy of VAD significantly impacts downstream tasks such as automatic speech recognition and noise-suppressed real-time communication, especially under low signal-to-noise ratio conditions where speech boundaries become blurred and noise interference intensifies. However, in noisy environments, existing methods still face challenges in handling fine-grained local features and long-range temporal dependencies. Furthermore, due to coarticulation effects, gradual energy decay, and transient noise, decision boundaries in VAD often remain unclear. To overcome these limitations, the authors propose a robust and efficient self-supervised distillation framework for real-time VAD. Experiments on the PTDB-TUG (PTDB-TUG: Pitch Tracking Database from Graz University of Technology) corpus demonstrate that the method achieves over 95% accuracy and F1 score even at a signal-to-noise ratio of 0 dB, outperforming state-of-the-art models in both robustness and efficiency.
Zitieren
APA
Lyu, Z., Xu, C., Cai, J., Jiang, Y., & Xiao, N. (2026). Semantic-Aware Self-Supervised Distillation for Efficient Real-Time Voice Activity Detection Systems. International Journal on Semantic Web and Information Systems, 22(1), 1-18. https://doi.org/10.4018/ijswis.405058
Vancouver
Lyu Z, Xu C, Cai J, Jiang Y, Xiao N. Semantic-Aware Self-Supervised Distillation for Efficient Real-Time Voice Activity Detection Systems. International Journal on Semantic Web and Information Systems. 2026;22(1):1-18. doi:10.4018/ijswis.405058
BibTeX
@article{Lyu2026,
author = {Lyu, Zhetao and Xu, Chenbin and Cai, Jianling and Jiang, Yutao and Xiao, Naijia},
title = {Semantic-Aware Self-Supervised Distillation for Efficient Real-Time Voice Activity Detection Systems},
journal = {International Journal on Semantic Web and Information Systems},
year = {2026},
volume = {22},
number = {1},
pages = {1--18},
doi = {10.4018/ijswis.405058}
}
EUVIMED