A multilinear (tensor) framework for HRTF analysis and synthesis

This paper introduces a multilinear (tensor) framework for the analysis and synthesis of the head-related transfer function (HRTF). The HRTF is the result of the confluence of two factors, sound location and person (anatomy). Our multilinear modeling technique employs a tensor extension of the conventional matrix singular value decomposition (SVD), known as the N-mode SVD [1] which explicitly represents the HRTF in terms of its constituent factors. Anatomical data is mapped to our multilinear HRTF model using regression. This mapping defines a data-driven model capable of producing different personalized HRTFs from easily obtained anatomical measurements. We show that our approach yields objectively superior results to those of a mapping based on principle components analysis (PCA).