Classification of humpback whale vocalizations using a self-organizing neural network

Describes a system for classifying vocalizations of humpback whales based on a source-filter model of sound production combined with a self-organizing feature map. Individual vocalizations were characterized in terms of their pitch, duration, noisiness, and formant structure using a combination of linear prediction, cepstral processing, and manual measurements. Vectors characterizing a sample of 242 sounds were then classified using a self-organizing feature map. The neural network partitioned vocalizations into categories that matched perceptually based classifications.