Characterizing and classifying humpback whale (Megaptera novaeangliae) song units.

Humpback whales, Megaptera novaeangliae, are one of the most recognizable and investigated marine mammals. However, little progress has been made in automatically distinguishing and classifying individual units of their song. A Matlab script has been developed to characterize the different song units and to apply the appropriate statistics to separate and categorize each unit. The Matlab program measures 48 parameters from each song unit. The songs were recorded by a swimmer snorkeling above vocalizing humpbacks in the waters off Maui, HI. A 16 bit, digital tape recorder with the automatic gain control disabled and a sample rate of 44.1 kHz was used to record songs from different whales. The swimmer determined the range of the whale using a portable handheld fathometer. Singing whales typically suspended themselves in the water column at depths varying from 15 to 30 m, which was contingent on the bottom depth. Song units were separated into distinct categories using a principle component analysis (PCA) based on the 48 parameters describing each unit. A classification algorithm was then developed based on the categories determined by the PCA. The classifier will be integrated into a modified version of the real‐time Odontocete call classification algorithm (ROCCA) for future analyzes.