Customization of head-related impulse response via two-dimension common factor decomposition and sampled measurements

A method based on subject-dependent impulse response extraction is proposed for the customization of head-related impulse response. In the training step, a two-dimension common factor decomposition algorithm is applied to train a set of direction-dependent impulse responses which are common for all subjects. A subject-dependent impulse response is extracted simultaneously for each subject to capture the subject-dependent information. In the customization step, the subject-dependent impulse response of a target subject is extracted from several head-related impulse response measurements of the subject. The extracted subject-dependent impulse response is then convolved with the trained direction-dependent impulses to construct all head-related impulse responses for the target subject. It is shown that with head-related impulse responses measured at a few directions for a target subject, head-related impulse responses at all trained directions can be customized with fairly low distortion.