Computing the personalized HRTFs based on weighted anthropometric parameters matching

Individual head-related transfer functions (HRTFs) are usually necessary for spatial sound field reconstruction and perception, which is important in many applications. However, measuring the personal HRTFs for different users is extremely resource consuming and impractical in real applications. An economic way is to compute the personalized HRTFs based on some existing HRTFs database by matching the anthropometric parameters between the target user and the subjects in the database. This paper proposes an improved method for computing the personalized HRTFs based on a public HRTF database, i.e., the CIPIC HRTF database. Instead of using the Euclidean distance for the ear parameters matching as proposed by Zoktin, a weighting scheme based on the statistical independency among different ear parameters was applied for the matching between the target user and the subjects in the database. The effectiveness of the proposed method, in comparison with that proposed by Zoktin, was evaluated with a cross validation method, i.e., the Leave-One-Out Cross-Validation. A set of sound source localization experiments using the HRTF-based virtual audio playback was also conducted with 5 subjects. Results show that the method based on weighted ear-parameters-matching obtained better performance in sound source localization.