HRTFformer: A Spatially-Aware Transformer for Individual HRTF Upsampling in Immersive Audio Rendering
Researchers have developed a new transformer-based architecture called HRTFformer to improve the quality of upsampled Head-Related Transfer Functions (HRTFs). This model uses attention mechanisms to better capture spatial correlations across the HRTF sphere, addressing limitations of previous machine learning approaches in preserving local spatial variations. By incorporating a neighbor dissimilarity loss for magnitude smoothness, HRTFformer achieves more realistic and accurate HRTF reconstruction from sparse measurements, outperforming existing methods in objective and perceptual evaluations. AI
IMPACT Enhances realism in immersive audio by improving HRTF reconstruction, potentially impacting VR/AR and gaming experiences.