Detection of cetacean and fish sounds using normalized summation of harmonics and spectrogram masking

Abstract

There is a growing interest in the ability to detect and classify animal vocalizations in large scale bioacoustic databases for the purposes of conservation and research. To aid in this, two methods are proposed for the quick and accurate detection of harmonic cetacean and fish vocalizations: Normalized summation of sound harmonics and spectrogram masking. These methods utilize a normalization scheme that enables robust performance, achieving 30% more precision and recall than traditional spectrogram cross correlation in the presence of wideband noise and low signal-to-noise ratios. The proposed methods also perform up to 135 times faster than spectrogram cross correlation.