An artificial neural network for studying binaural sound localization

A three-layer neural network is used to solve the problem of extracting relative azimuth and elevation positional information from signals detected by two directional receivers that are spatially separate. This is analogous to the ability of owls to localize the position of sound sources based solely on the properties of the signals reaching the two ears. Although the network implemented does not require any specific knowledge about acoustical parameters or propagation properties, a simple 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. The network successfully transforms phase and intensity differences of simulated acoustical signals into relative azimuth and elevation consistent with the simulated model