This research paper presents a new anthropometry-based method to personalize head-related transfer functions (HRTFs) using manifold learning in both azimuth and elevation angles with a single nonlinear regression model. The authors, from Unicamp and Redmond Microsoft Research, propose a graph construction procedure for learning a single Isomap manifold for all subjects that incorporates important prior knowledge of spatial audio to exploit the correlation existing among HRTFs across individuals, directions and ears.
The obtained results show that Isomap has a better performance and less variability than PCA as measured by the mean spectral distortion (MSD) with 95% confidence intervals. In addition, the results put in evidence that Isomap can capture high-frequency cues from intraconic HRTFs where PCA does not.
F. Grijalva, L. Martini, D. Florencio and S. Goldenstein, “A Manifold Learning Approach for Personalizing HRTFs from Anthropometric Features” in IEEE/ACM Transactions on Audio, Speech, and Language Processing, vol. 24, no. 3, pp. 559-570, March 2016. doi: 10.1109/TASLP.2016.2517565