Detection on the presence and frequency use pattern of cetacean tonal sound

Abstract

Passive acoustic monitoring (PAM) had already been proved to assess the presence of many cetacean species successfully. However, the continuous recording makes the manual data analysis difficult. In the present study, an automatic detection algorithm was developed for tonal sounds produced by Indo-Pacific humpback dolphins. The algorithm included a tonal sound detector which found the local spectral peaks and sampled the dominant frequencies every 5.3 ms. A noise exclusion process was used to exclude the spectral peaks with wide -3 dB bandwidth. After filtering the isolated frequency points within specific duration and frequency range, the adopted frequencies of tonal sounds could be obtained. The result showed the algorithm had 70% correct detection and 2.8% false positive based on each 1 sec time bin in 10 field recordings. The first to third quartile of adopted frequencies showed significant difference with those extracted manually, but the differences were only 245-489 Hz in average. The current algorithm performed considerably faster than real time. In the future, it can be applied as a first step in a real-time monitoring.