Technical and perceptual issues on head-related transfer functions sets for use in binaural synthesis

Head-related transfer functions (HRTFs) are the filters used in binaural synthesis that provide the necessary cues to localize virtual sound sources in virtual 3D space. HRTFs are recorded at the ears of either humans or dummy-heads and are usually implemented as non-individual filters. The success of binaural synthesis in providing a realistic and/or convincing virtual sonic experience relies largely on technical and perceptual issues inherent to the HRTFs and their use, particularly on how well the spectral features of the HRTFs used match those of the listener.

This PhD thesis focused on identifying the spectral features that cue sound localization in the mid-sagittal plane. Listening tests were conducted from which groups of individual and non-individual HRTFs that evoked the same direction were obtained. Analysis of these HRTFs suggested that the relevant spectral cues covered a broad range of frequencies. Parameters describing the spectral features were computed. Those parameters were seen to span ranges of values and frequencies and still evoke the same directions. These ranges seemed to be individual, ruling out the possibility of a unique set of non-individual HRTFs that overperforms current ones for a large sample of listeners. However, the spectral features and parameters identified could be potentially used in individual calibration procedures of 3D sound systems.

The PhD study also investigated other aspects of binaural technology which are relevant for the success of binaural synthesis, such as HRTFs measurement procedures, minimum-phase decomposition methods used with measured HRTFs and the audibility of high Q-factor all-pass sections from HRTFs. Sound source localization ability in human listeners and the possibility of sound quality degradation when spectrally non-matching HRTFs are used in binaural synthesis were also explored.