Individualization Methods Of Head-related Transfer Functions For Virtual Three-dimensional Sounds

Auditory display, especially three-dimensional (3D) auditory display, is becoming more and more important in a virtual reality system. This dissertation focuses on individualization of Head-Related Transfer Function (HRTF), which is one of the core technologies to generate virtual 3D sounds.First, the existing studies related to 3D virtual sounds based on HRTFs are reviewed. The necessity of HRTF individualization is discussed. Six potential solutions to HRTF individualization are summarized, among which HRTF individualization according to morphological features is considered simple and feasible. Second, error analysis is done for the Principal Components Analysis (PCA) and minimum-phase reconstruction method. The number of principal components for the model to meet a specific requirement is investigated. Third, an improved HRTF individualization method is validated, which adopts basis functions possessing the position information. The results indicate that the theoretical reconstruction error with the improved method could be significantly reduced. Fourth, a correlation analysis shows that the anthropometric measurements significantly correlated with HRTFs are from head, ear, and torso. The anthropometric measurements of head and torso have dominant effects on HRTFs at the low sampled frequencies while those of ears have dominant effects on HRTFs at the high sampled frequencies. Fifth, a weighted correlation method (WCM) is proposed to identify key anthropometric measurements (KAMs) for HRTF individualization. Compared with averaged HRTFs, the spectral distortion of the HRTFs individualized by these KAMs could be significantly reduced. Furthermore, by borrowing the sizing idea from anthropometry, a grouping method is proposed for partial HRTF individualization. By this method, a user can choose a sonic product as to choose a pair of shoes according to his/her own"size". Finally, a new HRTF tuning method based on subjective position perception is proposed to integrate subjective perception into HRTF spectral characteristics. To some extent, it compensates the insufficient consideration of subjective perception in the philosophy of traditional HRTFs. Experiment results show that the subjects could perceive the target directions after several trials of tuning.In summary, three solutions to HRTF individualization are studied in this dissertation: HRTF individualization based on anthropometry, partial individualization by population grouping, and HRTF tuning based on subjective position perception.