MARINE MAMMAL CALL DISCRIMINATION USING ARTIFICIAL NEURAL NETWORKS

Recent work has applied a linear spectrogram correlator filter (SCF) t o detect bowhead whale (Balaena mysticetus) song notes, outperforming both a time-series-matched filter and a hidden Markov model. The metho d relies on an empirical weighting matrix. An artificial neural net (A NN) may be better yet, since it offers two advantages; (i) the equival ent weighting matrix is determined by training and can converge to a m ore optimal solution and (ii) an ANN isa nonlinear estimator and can e mbody more sophisticated responses. A three-layer feed-forward ANN is ideally suited to this application and has been implemented on 1475 so unds, of which 54% were used for training and 46% kept as ''unseen'' t est data. The trained ANN error rate was 1.5%, a twofold improvement o ver previous methods. It is shown that ANN hidden neurons can be inter rogated to reveal the operating paradigm developed during training. Th e function of each of these neurons can be determined in terms of spec trographic features of the training calls. Furthermore, the operating paradigm can be controlled and training time reduced by assigning spec ific recognition tasks to hidden neurons prior to training, rather tha n initiating training with randomized weights. The ANN is compared to the SCF and the role of the ''hidden'' neurons and equivalent weightin g matrices are discussed.