Integration of real-time odontocete call classification algorithm into PAMGUARD signal processing software.

Abstract

Real-time odontocete call classification algorithm (ROCCA) is a tool for real-time acoustic species identification of delphinid whistles. Introduced in 2006 as MATLAB-based software, ROCCA is currently being incorporated into PAMGUARD, a freely-available, open source software package. ROCCA provides automated extraction of whistle contours from a spectrogram. It measures 54 whistle contour features including frequencies, slopes, duration, and variables related to the positions of inflection points and steps. ROCCA currently classifies whistles of seven species and one genus: Globicephala macrorhynchus, Pseudorca crassidens, Steno bredanensis, Stenella attenuata, S. coeruleoalba, S. longirostris, Tursiops truncatus, and Delphinus species. The classifier is a Random Forest trained on 2231 whistles collected over six cruises and 7 years in the eastern tropical Pacific Ocean. The original ROCCA classifier used a combination of discriminant function analysis and CART algorithms on 13 whistle contour features for an overall correct classification score of 35%, which was significantly greater than random (12%). The current Random Forest scheme, trained on 54 whistle contour features, yields an overall correct classification score of 62%. Feedback from at-sea beta testing has been incorporated into the latest version of ROCCA. Additional species, automated detection, and alternate classification schemes are being explored to extend ROCCA into different geographic areas with greater accuracy.