Head‐related transfer function compression using Legendre polynomials and autoregressive model.

Abstract

Head‐related transfer functions (HRTFs) are generally large datasets and thus a constraint for real‐time applications. We propose a method to reduce redundancy and compress the datasets. In this method, HRTFs are transformed into time domain head‐related impulse responses (HRIRs) and compressed by conversion into autoregressive (AR) filters. The AR coefficients are calculated using Prony’s method and the order is determined using the minimum eigenvalue of the HRIR covariance matrix. Such filters are specified by a few coefficients which can generate the full HRIRs. Next, Legendre polynomials (LPs) are used to compress the AR filter coefficients. LPs are derived on the sphere and form an orthonormal basis set for spherical functions. (Higher‐order LPs capture increasingly fine spatial detail; the number of LPs needed to represent an HRTF, therefore, is indicative of its spatial complexity.) The results indicate that compression ratios can exceed 95% while maintaining an error of less than 10% in the recovered HRTFs. [Work funded by the EU framework program 7 under Contract No. IST 215370.]