Soft-output signal detection for cetacean vocalizations using spectral entropy, k-means clustering and the continuous wavelet transform

Abstract

Underwater passive acoustic monitoring systems record many hours of audio data for marine research, making fast and reliable non-causal signal detection paramount. Such detectors assist in reducing the amount of labor required for signal annotations, which often contain large portions devoid of signals.Cetacean vocalization detection based on spectral entropy is investigated as a means of vocalization discovery. Previous techniques using spectral entropy mostly consider time–frequency enhancement of the entropy measure, and utilize the short time Fourier transform (STFT) as its time–frequency (TF) decomposition. Spectral entropy methods also requires the user to set a detection threshold manually, which call for knowledge of the produced entropy measures.This paper considers median filtering as a simple, effective way to provide temporal stabilization to the entropy measure, and considers the continuous wavelet transform (CWT) as an alternative TF decomposition. K-means clustering is used to determine the threshold required to accurately separate the signal/no-signal entropy measures, resulting in a one-dimensional, two-class classification problem. The class means are used to perform pseudo-probabilistic soft class assignment, which is a useful metric in algorithmic development. The effect of median filtering, signal-to-noise ratio and the chosen TF decomposition are investigated.The accuracy and specificity measures of the proposed detection technique are simulated using a pulsed frequency modulated sweep, corrupted by a sample of ocean noise. The results show that median filtering is particularly effective for low signal-to-noise ratios. Both the STFT and CWT prove to be effective TF analyses for signal detection purposes, each presenting with different advantages and drawbacks. The simulated results provide insight into configuring the proposed detector, which is compared to a conventional STFT-based spectral entropy detector using manually annotated humpback whale (Megaptera novaeangliae) songs recorded in False Bay, South Africa, July2021.The proposed method shows a significant improvement in detection accuracy and specificity, while also providing a more interpretable detection threshold setting via soft class assignment, providing a detector for use in development of adaptive algorithms.