Evaluating binaural rendering against stereo mixes is frequently confounded by "content bias," where listeners' inherent musical preferences obscure spatial quality assessments. To address this, we propose an interpretable predictive model utilizing a pairwise differential approach (Delta Strategy) and a dimension-wise attention neural network. The model achieves a competitive sign accuracy of 68.4%, outperforming traditional baselines. Crucially, the attention mechanism provides retrospective interpretability, revealing fundamental acoustic trade-offs in spatial upmixing: aggressive decorrelation for image widening compromises localization precision and timbral fullness, whereas successful externalization heavily depends on mid-side energy redistribution. This framework offers a robust evaluation tool for spatial algorithms and actionable psychoacoustic guidance for immersive audio production.
Designing interactive music systems in Virtual Reality (VR) requires balancing intuitive entry points with expressive depth, yet it remains unclear how domain-specific knowledge (Music Expertise) and medium-specific experience (VR Familiarity) distinctly shape the user experience within these environments. This paper investigates how user expertise impacts engagement with an interactive VR music experience. We conducted a mixed-methods study with 32 participants, categorized by these two factors, to systematically evaluate their influence on perceived usability, immersion, and interaction behavior. Results indicate that Music Expertise significantly enhanced perceived usability, whereas VR Familiarity had no significant effect. Perceived immersion was reported as universally high across all groups, regardless of background. Behavioral data revealed distinct engagement patterns: Experts and VR-familiar users focused more on 6DoF spatial mixing controls, while novices required significantly more time and physical exploration. These findings suggest that for creative VR tools, domain knowledge is a stronger predictor of usability than technical fluency. We discuss the success of a ‘Low Floor, High Ceiling, and Wide Walls’ design and propose critical design implications for onboarding, interaction metaphors, and aligning user intent in embodied music systems.
The perceptual evaluation of spatial and immersive audio systems commonly relies on listening tests, where the role of listener-related factors is often treated as secondary. While previous studies have shown that listener expertise can influence performance in virtual audio tasks, this has not been systematically investigated in more complex mixed real–virtual and dynamic listening scenarios. This study examines the role of listener background in a six-degrees-of-freedom (6DoF) spatial detection task involving virtual and real sound sources. Eighteen participants identified the presence of a virtual speech source among concurrent targets and distractors while freely navigating a loudspeaker-based scene. Listener background was characterised by years of musical training and self-reported experience with spatial audio technologies, used to categorise participants as expert or naïve. Results show above-chance performance, with reduced accuracy in spatially adjacent conditions. Listeners with greater musical training and spatial audio experience achieved higher percent-correct scores. These findings are consistent with prior work on listener-dependent localisation performance, and extend them to a 6DoF mixed real–virtual context. The results highlight the importance of explicitly considering and reporting participant expertise in the design, analysis, and interpretation of spatial audio perception studies.
Several studies suggest that singers adapt their vocal production to room acoustics, and virtual reality (VR) has increasingly been used to investigate such interactions under controlled conditions. However, questions remain regarding the ecological validity of virtual acoustic environments for studying musicians’ behavior. While prior research has primarily focused on solo singers, the present study explores the impact of acoustic modality (real vs. virtual) on choral performance. A professional four-singer ensemble performed five different choral pieces across five acoustic conditions. Recordings were conducted both in situ, within different spaces of a church, and under corresponding virtual acoustic simulations using auralization techniques. Acoustic and physiological data were collected using close microphones and electroglottography, while subjective perceptions were assessed through questionnaires. Comparative analyses between real and virtual conditions aim to examine how acoustic modality (real or virtual) influences singers’ musical and physiological adaptations, as well as their subjective perceptions.
In head-related transfer functions (HRTFs), spectral cues due to the individual pinna geometry are known to contribute to elevation perception and externalization. The pinna component of an HRTF is referred to as a pinna-related transfer function (PRTF). Some headphone concepts aim to excite individual PRTF cues by placing the headphone transducer away from the traditional position on the interaural axis, e.g. tilted in front of the pinna. However, it is not clear to which extent the individual PRTF is preserved when the pinna is placed inside a headphone cup enclosed by a baffle and a cushion. In this study, multiple prototype setups successively approximating a headphone cup and allowing for variable transducer positions are analyzed using a set of silicone pinna replicas. PRTF perturbations are analyzed in near field measurements and the impact of headphone cup acoustics is discussed. Based on the observation that the perturbations are systematic, an equalization scheme restoring the free field PRTF based on the median of measurements with several pinnae is proposed.
Virtual, augmented, and mixed reality experiences are becoming more commonplace as consumer-grade devices proliferate. Head-Related Transfer Functions (HRTFs) are used to create realistic spatial audio in virtual and augmented environments. Mathematically, HRTFs represent solutions to acoustic boundary-value scattering problems governed by the Helmholtz equation. Neural operators are neural networks designed to learn the solutions of partial differential equations (PDEs). The present work proposes an operator-learning framework based on the Deep Operator Network (DeepONet) for individualized HRTF prediction. By implementing a non-uniform sampling strategy for 3-D head meshes and data compression along the frequency axis, the framework achieves high-fidelity predictions while reducing data dimensionality. Our method shows low log-spectral distortion, generalizes to unseen spatial grids, and infers an entire head’s HRTF field in ~0.3 seconds. Objective evaluations demonstrate the framework's effectiveness in personalization and spatial interpolation. Furthermore, robust performance on unseen subjects and coordinates highlights the model's generalization capability, offering a computationally efficient alternative for HRTFs personalization.
Accurate loudness estimation is essential for audio production, quality control, and loudness compliance, but no established recommendation exists for binaural playback over headphones. This paper investigates the influence of binauralizers and HRTF processing on objective loudness estimation for binauralized Ambisonics content. Two experiments were conducted using 163 Ambisonics clips binauralized with two open-source renderers and three HRTF sets under three HRTF preprocessing conditions. Objective loudness metrics were compared against ground truth loudness data derived from 7.1+4 loudspeaker feeds according to ITU-R BS.1770. Results reveal small to moderate differences in Integrated Loudness and larger differences in the True Peak values between the evaluated binauralizers, and that diffuse-field equalization can effectively eliminate loudness and True Peak differences across binauralizers and across sets of HRTFs. The findings can help to better predict and ensure loudness compliance in binauralized audio consumption in XR and gaming, especially when importing 3rd-party HRTFs is supported.
Head-Related Transfer Functions (HRTFs) are commonly measured at the blocked ear canal entrance, assuming that the ear canal transfer function is direction-independent. While this assumption holds well at low and mid frequencies, its validity at high frequencies has been questioned. A recent pilot study on a single pair of 3D-printed ear replicas found evidence of directional effects above 9 kHz, but was limited in scope. This study extends that work using 3D-printed ear replicas of ten subjects from the IHA database, mounted on a dummy head. Ear canal transfer functions were measured across a full spherical grid of 1944 incidence angles. Results reveal significant directional variability above 6–7 kHz, with standard deviations of 6 –8 dB at resonant frequencies. High measurement repeatability confirms these are genuine directional effects rather than measurement artifacts. The directional behavior is consistently observed across all subjects and appears linked to the second and higher ear canal resonances. These findings suggest that the current state of the art blocked-canal HRTF measurements may omit spatially relevant spectral information above 7 kHz.
Head-related transfer functions (HRTFs) underpin spatial hearing in virtual and augmented reality systems. Whilst individual HRTFs capture listener-specific morphology, their practical limitations have led to widespread use of generic HRTFs and growing interest in synthetic approaches. Yet their relative perceptual impact remains rarely compared within a single study. In this study, twenty listeners completed two virtual reality sound localisation experiments with complementary subsets of interleaved HRTF conditions enabling within-subject comParison of five conditions: individually measured, KEMAR, randomly selected non-individual measured, high-resolution scan-based synthetic and photogrammetry-based synthetic HRTFs. Test–retest stability of the individually measured baseline across sessions supported pooling across experiments and attributing differences to perceptual rather than session effects. Across HRTF conditions, lateral localisation metrics were largely insensitive to HRTF type, whereas polar-domain metrics and confusion rates showed strong HRTF dependence. Random HRTFs outperformed KEMAR on several polar metrics. High-resolution synthetic HRTFs matched individual measured performance, whilst photogrammetry-based synthetic HRTFs, alongside KEMAR, showed the greatest degradation. These findings clarify practical choices for non-individual baselines and highlight the importance of mesh resolution when using numerical synthesis for elevation-dependent localisation tasks.
