Spline-based Head-Related Transfer Function compression

Krueger T, Villegas J. Spline-based Head-Related Transfer Function compression. In: AES Show 2026 Nashville. Audio Engineering Society; 2026.pp. 42:1-8. Available from: https://aes.org/publications/elibrary-page/?id=23473

@inproceedings{Krueger2026_23473,
  author = {Krueger, Tom and Villegas, Julián},
  title = {{Spline-based Head-Related Transfer Function compression}},
  booktitle = {AES Show 2026 Nashville},
  pages = {42:1-8},
  url = {https://aes.org/publications/elibrary-page/?id=23473},
  year = {2026},
  month = oct,
  publisher = {Audio Engineering Society},
  address = {Gaylord Opryland Resort}
}

TY  - CPAPER
TI  - Spline-based Head-Related Transfer Function compression
AU  - Krueger, Tom
AU  - Villegas, Julián
T2  - AES Show 2026 Nashville
SP  - 1
EP  - 8
M1  - Paper 42
PY  - 2026
DA  - 2026/10/30
UR  - https://aes.org/publications/elibrary-page/?id=23473
PB  - Audio Engineering Society
CY  - Gaylord Opryland Resort
LA  - en
AB  - Head-Related Transfer Functions (HRTFs) contain the spatial cues needed for immersive audio, but their size and the difficulty of personalizing them motivate compact, editable representations. We present a spline-based HRTF compression method that represents the magnitude spectrum by 43 control points per channel. One per Equivalent Rectangular Band (ERB) plus a high-frequency anchor, yielding a standardized format and achieving roughly 3:0:1 compression. Spectra are reconstructed with the Akimas method, a parametric method, and with a hybrid method that selects the best spline per segment, as an upper bound. Across 302 subjects from the SONICOM database, the mean spectral distortion achieved is 3:94 dB and 3:55 dB for Akimas and the hybrid method, respectively. A localization experiment with 18 participants, using a worst-case subject, found both reconstructions to be practically equivalent to the original HRTFs in polar angular error. In quadrant error, the equivalence held for Akimas method but not for the hybrid. These results indicate that preserving the most prominent spectral feature per ERB is sufficient to retain cues needed for localization.
ER  -