Evaluation of techniques for navigation of higher-order ambisonics

Abstract

Metrics are presented that assess spectral coloration and localization errors incurred by navigational techniques for higher-order ambisonics. Previous studies on the coloration induced by such navigational techniques have been largely qualitative, and the accuracy of previously-used localization models in this context is unclear. The presented metrics are applied in numerical simulations of navigation over a range of translation distances. Coloration is predicted using an auditory filter bank to compute the spectral energy differences between the test and reference signals in critical bands, and localization is predicted using a precedence-effect-based localization model. Coloration and localization errors are also measured through corresponding binaural-synthesis-based listening tests, wherein subjects are first asked to rate the induced coloration relative to reference and low-pass-filtered “anchor“ signals, and subsequently judge source position. Relationships are drawn between the metrics and the results of the listening tests in order to validate the predictive capabilities of the metrics.

Models for evaluating navigational techniques for higher-order ambisonics

“…Winter et al (2014) evaluated the localization accuracy of a plane-wave-based translation technique (Schultz and Spors, 2013) using a binaural localization model (Dietz et al, 2011) to predict perceived localization. Tylka and Choueiri (2015) compared the localization errors incurred by various translation techniques 1 using the velocity and energy localization vectors developed by Gerzon (1992). However, this analysis neglected the precedence effect, which is expected to play an important role in the context of sound field navigation, as an accurate virtual translation of the listener necessarily involves direction-dependent time shifting of incident signals.…”

Performance of Linear Extrapolation Methods for Virtual Sound Field Navigation

“…Perhaps the most intuitive navigational method is the virtual ambisonics method, as described by Tylka and Choueiri [9,Sec. 3.1].…”

“…The localization properties of this method were explored by Winter et al [16], who showed that the range over which accurate localization is achieved increases with HOA expansion order and that increasing the number of planewave-expansion terms beyond critically-sampled does not improve localization. More recently, Tylka and Choueiri [9] evaluated the localization errors using the velocity and energy localization vectors developed by Gerzon [17], although the perceptual relevance of the findings of this study are limited since the analysis does not take into account the precedence effect.…”

“…The signal amplitudes, plane-wave directions, and times-of-arrival for all wavelets are fed into the original energy vector model of Stitt et al to produce a single predicted source localization vector. 9 In their model, Stitt et al predict localization as a weighted average of position vectors for each of a finite set of discrete sources. Perceptual weights are first computed for each source based on its time-of-arrival to and amplitude at the listening position.…”

“…62]. 9 More precisely, this operation is performed in several frequency bands, and a weighted average localization direction is computed using signal energies in each band as weights [43, Sec. 2.A.i].…”