Statistical analysis of head related transfer function data

• Published in: The Journal of the Acoustical Society of America 2013-05, Vol.133 (5_Supplement), p.3281-3281
• Main Authors: Luo, Yuancheng, Zotkin, Dmitry N., Duraiswami, Ramani
• Format: Article
• Language: English
• ISSN: 0001-4966; 1520-8524
• DOI: 10.1121/1.4805368

The head related transfer function (HRTF) is a function that characterizes the response of a given individual to sound from a particular location in an egocentric coordinate system. The range dependence is often neglected, and the HRTF is approximated as a function of frequency and direction, H(θ,φ,ω). The HRTF displays considerable inter-personal variability, and a major open problem is the development of a generative model for the HRTF from anthropometry. Further, the sampling used in measuring HRTF data varies widely from database to database, and moreover often there are no measurements for elevations below the subject. This raises associated questions of optimal sampling, interpolation, hole-filling, and others. In this work, we model the HRTF via a non-parametric, data-driven, Gaussian Process Regression model. We develop efficient regression techniques to perform inference using this model on measured HRTF data. We then suggest methods for HRTF interpolation, HRTF extrapolation, feature extraction, sampling, and personalization. The methods are tested on the CIPIC database and results presented. [Partial NSF support is gratefully acknowledged.]