ABSTRACT
Artificial neural networks were trained to identify the location of virtual sound sources based on the information in head-related transfer functions (HRTFs) recorded from a human subject. The results were used to evaluate whether either a monaural representation of stimulus information or an interaural representation of stimulus information is adequate to mediate human spatial hearing performance. The simulated signals were filtered clicks presented from virtual speakers placed at 15° steps in azimuth and 18° steps in elevation. After the signals were passed through the HRTFs, quarter-octave spectra were computed. The inputs to the networks were the monaural spectra, the interaural difference spectrum, and/or the interaural time delay. Back propagation was used to train individual networks for each combination of stimulus information. Depending on the stimulus informa tion presented to the network, performance could be much worse or much better than that of the human subject. Overall, the results indicate that the interaural time delay in combination with either monaural or interaural spectral information is sufficient to produce performance comparable to that observed for human listeners.
