Marine animal sound classification

Abstract

Software was developed to measure characteristics of marine animal sounds (AcouStat). These measurements proved effective for classifying sounds in several contexts: identifying species, quantifying the repertoire of a single species, and identifying individuals. The sound measures included statistics for aggregate bandwidth, intensity, duration, amplitude modulation, frequency modulation, center frequency, and interactions among these variables. Classification analysis based on these measures suggests they adequately characterize the variability of bioacoustic signals for many problems. Correct classification to species was as high as 85%, and correct classification of dolphin whistles to individual was 90%.