| This work describes an alternative approach to delivering high spatial-quality binaural sound to users, through database matching. An algorithm was designed that pairs users to pre-measured Head-Related Transfer Functions (HRTF), based on a set of sparsely measured impulse responses. This work has significance in several fields. In education, it can improve communication between collaborating people located in different sites; in entertainment, it can result in more immersive experiences for listeners; in mission-critical applications, it can deliver enhanced and accurate reconstructions of one's auditory environment, and improve spatial awareness.;Several contributions were made through the course of this research. A large standardized repository of HRTFs was created from publicly available databases, as a means of improving the database matching system's performance. A new file format for storing this data was also defined, allowing for smooth navigation between sets, and easy expansion of the collection with new measurements. The length of the training data was further increased through original measurements, which resulted in the first publicly available database of repeated HRTFs on human participants. This database formed the basis for proposing an alternative method for HRTF dimensionality reduction through Linear Discriminant Analysis, as well as for exploring variations in measured HRTFs and their potential perceptual impact.;The effectiveness of the system was evaluated through a series of subjective evaluations, assessing spatial accuracy and user preference. Twenty people volunteered to participate in free-field and binaural localization tasks, as well as in HRTF evaluation studies. In general, analysis of the results indicated that the designed method, was successful in pairing users to the most appropriate HRTFs from a large repository of available sets. The effectiveness of the match was found to be related to its level of similarity to the query, as well as the expertise of the user, and was bound by the available data. |
