Sound-localisation performances using head-related transfer functions calculated from raw and improved photogrammetric scans (en)
* Presenting author
Abstract:
One important key for plausible binaural rendering is the usage of individual head-related transfer functions (HRTFs). In the past two decades, techniques for the numerical calculation of HRTFs have improved significantly. Yet, plausible results can be achieved only if an accurate 3D mesh of the listener is available. Photogrammetric reconstruction (PR) is a promising approach to acquire individual ear geometries for a wide audience but, especially in occluded areas, PR scans do not yield geometric accuracy. In this investigation, we conducted a localisation experiment comparing sound-localisation performances using individual acoustically measured HRTFs and numerically calculated HRTFs based on the following four geometry acquisition methods: reference laser 3D scan, PR-based acquisition with a mobile phone, PR-based geometry improved by a parametric pinna model (PPM), and PR-based geometry improved by a non-rigid registration (NRR) algorithm. The results from the experiment will be presented and the importance of NRR-based and PPM-based PR improvements will be discussed.