An artificial neural network for sound localization using binaural cues

A three-layer neural network is used to estimate the direction of a sound source from the signals detected by two directional, spatially separate receivers. Although the implemented system does not require any specific knowledge about acoustical parameters or propagation properties, a model of the acoustical environment is used to generate simulated data for training the network. The neural network is trained according to the multiple extended Kalman algorithm (MEKA), which provides fast convergence and does not require intervention for adjustment of the learning parameters. Lower bounds on estimation are computed and compared with simulations using the neural network. (C) 1996 Acoustical Society of America.