Blue whale calls classification using short-time Fourier and wavelet packet transforms and artificial neural network

Two new characterization methods based on the short-time Fourier and the wavelet packet transforms are proposed to classify blue whale calls. The vocalizations are divided into short-time overlapping segments before applying these transforms to each segment. Then, the feature vectors are constructed by computing the coefficient energies within two subbands in order to capture the AB phrase and D vocalization characteristics, respectively. Finally, a multilayer perceptron (MLP) is used to classify the vocalization into A, B and D classes. The proposed methods present high classification performance (86.25%) on the tested database.