Semantic Communication Framework for Anomalous Sound Detection
IEEE Communications Letters · 2026 · 30 · 3708-3712
DOI: 10.1109/lcomm.2026.3747590 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Communications Letters, 2026. Verfasst von Wenmin Luo und 3 weiteren Autor:innen. Die Arbeit nennt 17 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Key Research and Development Program Foundation of Zhejiang.
Abstract (Originalsprache)
Deep-learning-enabled semantic communication (DeepSC) optimizes the transmission of task-relevant information, making it uniquely suited for unsupervised, one-class anomalous sound detection (ASD). Motivated by this, we propose MR-DeepSC, a multi-scale and multi-resolution semantic communication framework. Because the system operates under a strict one-class unsupervised paradigm, MR-DeepSC redefines semantic preservation as the retention of multi-scale time-frequency structures rather than generic waveform fidelity. To accurately measure this latent structural integrity, we utilize Semantic Information Retention (SIR) and Semantic Vector Alignment (SVA) to quantify representation preservation via a frozen ASD model during evaluation. Experimental results demonstrate that MR-DeepSC retains over 95% of the Clean Audio benchmark performance across both the AUC and the pAUC metrics under an AWGN channel at 8 dB SNR on the DCASE 2020 development test partition.
Förderung
- National Natural Science Foundation of China
- Key Research and Development Program Foundation of Zhejiang
Zitieren
APA
Luo, W., Luo, S., Luo, Z., & Xu, W. (2026). Semantic Communication Framework for Anomalous Sound Detection. IEEE Communications Letters, 30, 3708-3712. https://doi.org/10.1109/lcomm.2026.3747590
Vancouver
Luo W, Luo S, Luo Z, Xu W. Semantic Communication Framework for Anomalous Sound Detection. IEEE Communications Letters. 2026;30:3708-3712. doi:10.1109/lcomm.2026.3747590
BibTeX
@article{Luo2026,
author = {Luo, Wenmin and Luo, Shuyun and Luo, Zhiyi and Xu, Weiqiang},
title = {Semantic Communication Framework for Anomalous Sound Detection},
journal = {IEEE Communications Letters},
year = {2026},
volume = {30},
pages = {3708--3712},
doi = {10.1109/lcomm.2026.3747590}
}
EUVIMED