Automatic template matching for classification of dolphin vocalizations

Whistle classification is a key step in many studies of dolphin vocalizations. Automatic whistles tracing algorithms have been developed but tracing errors such as breaks and outliers are usually unavoidable. Local variations of whistle contours occur even in whistles of the same type. In this paper, we describe a modified dynamic time warping (DTW) algorithm for dynamic non-linear matching. It exhibits a good performance in watching against a template whistle. The modifications to the basic DTW algorithm provide improved tolerance to noise and breaks in tracing. Together with automatic de-noising, this template matching is used to classify vocalizations of Indo-Pacific dolphins (Sousa chinensis). We believe this method can be applied for large scale analysis of whistles for species recognition, dolphin training and other dolphin studies.