Automatic Classification of Whistles Produced by Indo-Pacific Humpback Dolphins (Sousa chinensis)

A fast, robust technique is needed to facilitate studies of vocalisations by dolphins and other marine mammals such as whales in which large quantities of acoustic data are commonly generated. It is sometimes necessary to be able to describe whistle contours quantitatively, rather than simply looking at descriptors such as start frequency, maximum frequency, number of inflection points, etc. This is important when whistles are to be compared using an automated classification system, and is an essential component of a real-time, automated classification system for use with a raw data stream. In this paper we describe a rapid and robust high order polynomial curve fitting technique which extracts features in preparation for automated classification. We applied this method to classify natural vocalizations of Indo-Pacific humpback dolphins (Sousa chinensis). We believe the method will be widely applicable to bioacoustic studies involving FM acoustic signals in both underwater and in-air environments.