Jérôme Amouyal, David Rainton
Sound Localization in the Horizontal Plane: A Binaural Approach
Abstract:It is generally believed that humans localize sound in the horizontal plane using the Interaural Time Difference (ITD) and Interaural Amplitude Difference (IAD) between the signals received at the two ears. In this research the performance of an acoustic localization device composed of a front end binaural simulator and back end neural network classifier is evaluated.
First a small binaural database of speech, clapping, and music sounds was recorded using the ATR Head and Torso Simulator. Using the data, three different neural networks were trained to localize sounds using ITD, IAD and both ITD and IAD cues. Results presented indicate that the optimal window lengths for both ITD and IAD cue computation are of the order of 100ms. The network using both ITD and IAD cues outperformed those using just ITD or IAD information alone. Finally the construction of a parallel real time acoustic localization device is described.