Techniques for segregation and classification of several vocalizing sperm whales for AUV-based localization applications

Sperm whales are amongst the most vocally active cetaceans. A fuller knowledge of these whales and their clicking behavior can be obtained by creating a picture of the relative locations of sperm whales in a region of interest and also by tracking them individually in three dimensions. This paper reports a major step in this direction by the development of automatic techniques for segregation of several clicking individuals and to generate a de-interleaved time series of their individual vocalizations. We discuss methods to extract the clicks from hydrophone data and for individual classification based on various click parameters: time-delay between (hydrophones, inter-pulse-interval within a click, correlation with a reference, and the peak frequency. We present results of our techniques for automatic segregation and classification of several sperm whales using data collected from a 2-hydrophone array. These results demonstrate their suitability for a compact real-time implementation on AUVs.