Thomas J. Hayward
Naval Res. Lab., Washington, DC 20375-5350
A multiple-resolution, Bayesian statistical approach to classification [J. Acoust. Soc. Am. 96, 3312(A) (1994)] is tested on simulations of marine-mammal sounds received in an ocean waveguide. Training data for several types of marine-mammal vocalizations are obtained from both the Woods Hole Oceanographic Institution SOUND database and the Naval Research Laboratory Dual Use Acoustics Center (DUAC) database. Sound samples not in the training set are then presented to the classification algorithm in a simulation of sound emanating from multiple marine-mammal sources [J. Acoust. Soc. Am. 97, 3371(A) (1995)]. This simulation incorporates modeling of ocean acoustic propagation effects on the vocalization waveforms. When combined with event detection, the classification algorithm achieves near-real-time speed in classification of several types of marine-mammal sounds. [Work supported by ONR.]