Automatic detection and classification of cetacean tonal sounds from a long-term marine observatory

Abstract

Passive acoustic monitoring is effective for examining the temporal variation of cetacean occurrences through continuous recording. However, the interpretation of behavior variability remains difficult. Cetacean tonal sounds are believed to play an important role in communication. Changes in tonal sound usage can be indicative of changes in cetacean behavior. In this study, we applied a generalized tonal sound detector on recordings from the marine cable hosted observatory (MACHO) off Northeastern Taiwan. The tonal sound detector is a modified local-max detector, which examines the prominence of tonality in 2 ms time intervals. Detections were confirmed based on the short-time transient characteristic of tonal sounds. The burst pulse, fundamental frequency, and harmonic were separated during post-processing. The detection results can be used to analyze the frequency distribution and modulation of the tonal sounds. Following individual contour extractions, the tonal sounds were classified to examine the repertoire composition through duration and frequency normalization. An evident nighttime occurrence was identified for both the burst pulse and tonal sound. The mode frequency of tonal sound did not show a significant difference, but the percentage of overlapping tonal sounds, Shannon entropy of frequency distribution, and repertoire complexity varied significantly among diurnal cycles. The current framework of quantification on tonal sound modulation and repertoire complexity provides intuitive parameters for the variability of cetacean behavior, which can be used to examine the temporal variations of cetacean habitat use. The precision in distinguishing different species and behaviors may be increased by comparing the repertoire established through automatic classification.