Classification of bioacoustic time series utilizing pulse detection, time and frequency features and data fusion
FacultiesFakultät für Informatik
LicenseStandard (Fassung vom 03.05.2003)
Classifying the sounds of species is a fundamental challenge in the study of animal vocalizations. Most of these studies are based on manual inspection and labeling of sound spectra, which relies on agreement between human experts. In this study recorded songs of crickets (Grylloidea) from Thailand and Ecuador are analysed and classified automatically. For this, the locations of pulses are determined and different features from the time and frequency domain are extracted automatically from the time series. For the categorization of the sound patterns these different features are combined through data fusion, temporal fusion and decision fusion. Local features and global features of the sound patterns are distinguished. For the classification a fuzzy-k-nearest-neighbour classifier is used. This classifier scheme exhibits a large similarity to artificial neural networks, in particular to radial basis function neural networks. We present classification results for a data set of 28 different species.