Analysis of most prominent signal features of humpback whale (Megaptera Novaeangliae) vocalizations towards the goal of autonomous acoustic classification.

Abstract

Humpback whale vocalizations were recorded using hydrophones on glider systems off Alaska in January 2000, in Hawaii in February 2008, and in the Stellwagen Bank National Marine Sanctuary in October 2007 and July 2008. The vocalizations have been grouped into five call types based on the most prominent signal features. Only five call types are used because autonomous species classification relies on the most consistent and repeatable signal features rather than the full diverse range of humpback vocalizations. The five call types are upsweep (increasing frequency over time), downsweep (decreasing frequency over time), flute (increasing and decreasing frequency over time), tone (little or no change in frequency over time), and groan (commonly a social or feeding-related vocalization, frequently characterized by unstructured broadband sound). We present detailed statistical analyses of these call types including bandwidth, minimum and maximum frequency, duration, and slope. A comparative analysis across data sets shows the relative frequency of occurrence of each vocalization type and indicates the degree of temporal and geographic variation of Humpback vocalizations.