Modeling HRTF for Sound Localization in Normal Listeners and Bilateral Cochlear Implant Users

Mathematical models can be very useful for understanding complicated systems and for testing algorithms through simulation that would be difficult or expensive to implement. This dissertation presents a model that attempts to simulate the sound localization performance of persons using bilateral cochlear implants. The expectation is that this model could prove to be a useful tool in developing new signal processing algorithms for neural encoding strategies. The head related transfer function (HRTF) is a critical component of this model, and in the ideal case, provides the base characteristics of head shadow, torso and pinna effects. This defines the temporal, intensity and spectral cues that are important to sound localization. By building on the HRTF, a sound source localization model can be constructed. This model was first developed to simulate normal hearing persons and validated against published literature on HRTFs and localization. The model was then further developed to account for the differences in the signal pathway of the cochlear implant (CI) user due to sound processing effects, and the microphone location versus pinna and ear canal acoustics. Finally, the localization error calculated from the model for cochlear implant users was compared to published localization data obtained from these hearing impaired patients in order to validate the modified model. Results of the normal hearing model correlated closely with localization performance data published in the literature, with localization error of the model only slightly greater than that of normal hearing subjects. The cochlear implant population has a more broadly distributed range of localization error than that of the normal hearing population, and in addition, the mean error is significantly poorer. The performance of the cochlear implant model fell within the range of error reported in the research literature for cochlear implant users. This close correspondence with the published performance data suggests that the model developed in this dissertation provides a reasonably good approximation of sound source localization for normal hearing subject and persons with bilateral cochlear implants