Abstract
A model is presented for synthesizing Head-Related Transfer Functions (HRTFs) in three-dimensional audio reproduction. The spatial and temporal variations of HRTFs are separated, using the singular value decomposition (SVD). The order of the model can be reduced by using this method to yield significant enhancement of processing efficiency. Owing to the discrete nature of measured HRTFs, interpolation is performed on the spatial part of the model to result in finer resolutions. The HRTF at an arbitrary direction is represented as a weighted sum of the products of the spatial and temporal parts. Performance analysis of the algorithm reveals that the efficiency of this model considerably improves with increasing number of sound sources. Computational efficiency and storage requirement are assessed for HRTFs measured from a manikin. Subjective tests of sound source localization are also conducted. The results indicated the effectiveness of the proposed HRTF synthesis technique for both azimuth and elevation localization, without notable performance degradation.