Adaptive spectral subtraction for HF audio denoising in maritime intelligence applications

Sulaiman, Mohd Azman Hanif and Mohd Yassin, Ihsan and Mohd Shariff, Khairul Khaizi and Md Tap, Khairuddin and Mohd Nasushardin, Muhamad Haslimin and Saidin, Aiddil Fariz and Md Tahir, Nooritawati (2026) Adaptive spectral subtraction for HF audio denoising in maritime intelligence applications. Journal of Electrical and Electronic Systems Research (JEESR), 29 (1): 2. pp. 13-25. ISSN e-ISSN : 3030-640X
Identification Number (DOI): 10.24191/jeesr.v29i1.002
Abstract

The Royal Malaysian Navy (Tentera Laut Diraja Malaysia, TLDM) conducts continuous wideband radio-frequency surveillance across maritime operational areas using an array of high-frequency (HF) wideband antennas. These intercepts are severely degraded by channel noise, co-channel interference, and atmospheric propagation effects, which reduce signal quality and clarity. This paper presents a systematic evaluation of an adaptive spectral subtraction method for denoising these HF recordings. A corpus of 15 real-world HF recordings totaling 4,723.65 seconds was obtained from surveillance operations and processed using three parameter configurations to identify the most suitable setting. The objective was to design and implement a signal-cleaning procedure that suppresses unwanted noise while preserving the information-bearing signal. Root-mean-square (RMS) energy reduction, Pearson correlation, and log-spectral distance (LSD) were computed for each file-parameter combination, yielding 45 evaluation points. The adaptive spectral subtraction method with a noise percentile of 5% and a Savitzky-Golay smoothing window of 51 achieved the best balance between noise suppression and signal fidelity, with a mean correlation of 0.989 and a mean LSD of 3.92 dB across all noise categories.

Item Details
Edit Item
Edit Item
Downloads & Files
[thumbnail of 149105.pdf]
Text
149105.pdf
Download (1MB)
Indexing & Metrics