HRIR-Former: Grid-Free Time-Domain Reconstruction of Head-Related Impulse Responses with a Spatially Encoded Transformer
Researchers have developed HRIR-Former, a novel time-domain Transformer model designed to reconstruct Head-Related Impulse Responses (HRIRs) at arbitrary directions from sparse measurements. Unlike previous methods that often operate in the frequency domain or rely on minimum-phase assumptions, HRIR-Former reconstructs HRIRs directly in the time domain, avoiding potential degradation in temporal fidelity and spatial continuity. The model incorporates sinusoidal spatial features, a Conv1D refinement module, and auxiliary heads for interaural time and level differences, demonstrating improved performance on the SONICOM dataset compared to existing techniques. AI