VR Audio Study: High-Res Synthetic HRTFs Match Individualized Performance
Photogrammetry-based HRTFs lag behind, while random generic ones beat KEMAR.
In a study presented at the AES 2026 Conference on Audio for Virtual and Augmented Reality, researchers Pirard and Poole compared five Head-Related Transfer Function (HRTF) conditions across 19 listeners in two VR sound localisation experiments. The conditions included individually measured, KEMAR (standard dummy head), randomly selected non-individual measured, high-resolution scan-based synthetic, and photogrammetry-based synthetic HRTFs. Using within-subject interleaved trials and test-retest stability checks, the team isolated perceptual differences from session effects.
Results showed that lateral localisation metrics were largely insensitive to HRTF type, but polar-domain metrics and confusion rates varied strongly. Surprisingly, randomly selected non-individual HRTFs outperformed the widely used KEMAR baseline on several polar measures. High-resolution synthetic HRTFs (from structured-light scans) achieved performance on par with individually measured HRTFs. In contrast, photogrammetry-based synthetic HRTFs performed as poorly as KEMAR, highlighting that mesh resolution is critical for elevation-dependent localisation. The findings provide clear guidance for VR/AR developers: a well-made synthetic HRTF can substitute for individual measurement, but low-resolution methods degrade spatial audio quality.
- High-resolution synthetic HRTFs matched individually measured HRTF performance across polar and lateral metrics.
- Random non-individual HRTFs outperformed the industry-standard KEMAR dummy head on several elevation-sensitive tests.
- Photogrammetry-based synthetic HRTFs showed the worst degradation, on par with KEMAR, underscoring mesh resolution's importance for vertical localisation.
Why It Matters
VR/AR audio developers can now justify high-res synthetic HRTFs over expensive individualisation, while avoiding poor-performing photogrammetry or KEMAR baselines.