In the state-of-the art models of sagittal-plane sound localization, template head-related transfer functions (HRTFs) are used to reflect the listener's internal calibration of auditory space decoding, and thus determine the prediction quality. The effect of the template HRTFs has not yet been investigated directly. Here, a model was calibrated separately to two HRTF measurements of the same listeners and its predictions were compared to behavioral localization responses of these listeners obtained in three listening conditions: two acoustically measured HRTF sets (those used for model calibration), and an additional condition (unseen during the calibration) used to test the model's ability to generalize. We analyzed the quadrant error rates (QE) and local polar errors (PEs) from eight listeners. The predicted errors were similar in both calibration conditions and increased in the unseen condition. The quality of the predictions, however, varied significantly with the template, more for PE than for QE, slightly preferring one template over the other when predicting the unseen condition. Our findings suggest that small differences in HRTFs used for the template may influence the prediction quality, especially when applied to unseen listening conditions.
Binaural rendering is central to spatial audio reproduction via headphones and wearable devices, yet systematic evaluation of enhancement techniques remains methodologically inconsistent across the literature. This paper presents and applies a subjective evaluation methodology designed to consistently elicit five perceptual attributes of headphone-based spatial audio: externalization, elevation fidelity, front/back distinction, room tone, and spectral coloration. The methodology combines absolute judgment and relative comParison protocols, taking advantage of their complementary capabilities to capture both the absolute perceptual quality of individual rendering conditions, and the salience of perceived differences between them. It is applied in a controlled experiment comparing conventional HRTF-based binaural rendering against two enhancement variants that each superpose a masked spatially diffuse sound field component into the binaural output. Stimuli were spatialized across eleven azimuthal positions in the horizontal plane using a generic dummy-head dataset, and presented over closed-back headphones to participants. The results validate the proposed methodology as a tool for revealing perceptually relevant differences among binaural rendering conditions. Relative comParison tests reveal additional performance difference details between rendering methods. Both enhancement methods significantly improve externalization and front/back distinction relative to unenhanced binaural rendering, with the largest gains at lateral azimuths, and without a statistically significant increase in perceived spectral coloration beyond the baseline effect of HRTF filtering. However, the results do not indicate conclusively whether the binaural rendering methods examined here exhibit or mitigate a "spurious elevation" artifact associated with frontal sound presentation.
