Decreased 3D-sound spatialization accuracy caused by speech bandwidth limitation over commodity audio components

This paper studies the accuracy achievable in spatialized speech signals due to their limited bandwidth, in systems based on generic Head-Related Transfer Functions (HRTFs). The accuracy in the perception of sounds spatialized through HRTFs to emulate nine sound locations around the listener, at 0 elevation, are investigated. The tests contrast the localization accuracy for a speech segment and that for broadband noise, as perceived by two independent groups of 10 normal-hearing volunteers each. The sound localization perceived and reported by each subject is compared to the HRTF-emulated location, to define a localization error, in each case. The results are analyzed through a repeated measures, mixed-factorial design Analysis of Variance (ANOVA). The emulated source locations and the type of sounds are the independent variables, and the average perceived localization error for each case, the dependent variable. Since the spectrum of speech signals drive only sections of the complete HRTF frequency response for each location, the initial expectation is that they benefit less from the spectral-shaping process implemented by the HRTFs, thus resulting in a lower spatialization efficiency than the one for a full-spectrum noise signal. We study the measurement of these effects when 3D-sounds are delivered over commodity audio components, because of their widespread use in practical systems.