Generation of binaural information using cross ambiguity functions

Abstract

This paper describes a method to generate binaural information. First of all, the accuracy of a conventional linear interpolation was examined by using signal-to-deviation ratio (SDR) and spectrum distortion. It was found that the accuracy of left and right direction HRTFs (60−120°) did not deteriorate despite the presence of a wide directional interval (SDR ≂ 23 dB). However, the accuracy of front and back direction HRTFs (0−−60° and 120−180°) deteriorated as the interval broadened. For the shade-side ear’s HRTFs, particularly, it was confirmed that the accuracy of linear interpolation deteriorated dramatically (SDR≂8 dB). Therefore, a method was investigated for generating a shade-side ear’s HRTF by using cross ambiguity functions (CAFs) whose model is related to dichotic listening. A shade-side HRTF could be generated with about a 10-dB SDR from a sunny-side HRTF, while the SDR becomes about 8 dB for a conventional linear interpolation estimation method. In this paper, the independence of HRTFs is also discussed, using the singular value decomposition from the point of view that it is possible to generate a new vector from a linear combination of vectors. In this article the accuracy of a linear interpolation from a base of HRTFs by SDR and spectrum distortion is examined.