Sound localization with a neural network trained with the multiple extended Kalman algorithm

A three-layer neural network is used to solve the problem of extracting relative azimuth and elevation positional information from signals detected by two spatially separate, directional receivers. This is analogous to the ability of owls to localize the position of sound based solely on the properties of the signals reaching their ears. A simple model of the acoustical environment was used to generate simulated data for training the network. The neural network was trained using the multiple extended Kalman algorithm (MEKA). MEKA enabled the network to be trained without constant user intervention for adjustment of the critical parameters of the model