A hybrid computational framework for room-acoustic prediction and assessment:Room impulse response-based parameter analysis and HRTF-based auralization

During building design, acousticians must quickly and accurately communicate the acoustic consequences of architectural decisions to project stakeholders. Certain geometric choices can strongly affect room-acoustic performance and may be difficult or impossible to change once a room has been built; in the worst case, they can prevent a space from fulfilling its intended function altogether. To gain insight into the acoustical performance of spaces that have not yet been built, acousticians rely on statistical sound-decay formulations and room-acoustic simulation tools. Room-acoustic simulation tools used in the field are predominantly based on geometrical acoustics, which model sound as rays and therefore do not inherently capture wave phenomena such as diffraction and interference. This can lead to overlooked design consequences, for example diffraction around obstacles or low-frequency modal behavior. Wave-based solvers capture these effects but become exponentially more computationally expensive with increasing frequency, effectively limiting their practical use to the low-frequency range in realistic room scenarios. This dissertation responds by developing and evaluating a hybrid room-acoustic modeling framework that employs a wave-based solver at low frequencies and, at high frequencies, combines a geometrical specular component with a statistical non-specular component. The crossover frequency fc is chosen based on the Schroeder frequency. Above fc, the specular reflection component is computed using an image-source model, while the non-specular component is synthesized using an envelope-based approach that treats the scattered field as approximately isotropic (already in the early stage) and gradually transitions to a late-time diffusion-equation envelope. The overall framework is evaluated against a stand-alone wave-based reference using perceptually relevant room-acoustic parameters across different scenarios. This validation shows that the hybrid approach can reproduce relevant room acoustical parameters, such as EDT, T20, Ts, and BR, within perceptual thresholds across the investigated scenarios. The considered scenarios are compact and feature a relatively homogeneous distribution of material properties, conditions that promote diffuse-field-like decay behavior and therefore support the assumptions underlying the non-specular high-frequency part of the framework. Motivated by this assumption, a structured approach was developed to determine when classical single-decay statistical formulations remain valid under imperfectly diffuse conditions. Using geometrical sound-field simulations in cuboid rooms, enclosure compactness and boundary-property homogeneity were varied systematically. From the simulated room impulse responses, absorption and scattering coefficients were reconstructed from fitted exponential decay functions and compared to the mean area-weighted coefficients commonly assumed in single-decay statistical formulations. The results show that enclosure compactness and boundary-property homogeneity jointly determine validity: compact rooms with homogeneous (or balanced) boundary conditions align well with statistical predictions, whereas elongated rooms and absorption differences between parallel-wall sets produce direction-dependent decay and systematic deviations. Increased scattering improves sound-field mixing and relaxes these constraints. Perceptual relevance is assessed by mapping deviations to reverberation-time differences, showing that errors can exceed the just noticeable difference (JND) and reach roughly 6-8 x~JND when both compactness and boundary homogeneity are violated. These findings provide guidance on the applicability of single-decay statistical theory, and the structured framework can be used to diagnose the validity of diffuse-field assumptions across different room geometries and boundary-condition configurations. To enable auralization-based assessment, the hybrid framework is extended from monaural RIRs to binaural room impulse responses (BRIRs) by integrating head-related transfer functions (HRTFs) into each model component. HRTFs are interpolated from a high-resolution set for the specular image-source component, a spatially averaged HRTF is used for the non-specular part of the high-frequency range, and a plane-wave decomposition approach is applied to the wave-based solver. This integration raises a key question, as most virtual auditory display implementations rely on far-field HRTFs that assume distance independence and dependence only on incidence angle. The consequences of applying this far-field assumption under near-field conditions are examined, and two sources of mismatch are distinguished. The first consists of analytically addressable geometric effects due to the ears' displacement from the center of the head, and the distance-dependent mismatch between ear-centered and head-centered source directions (acoustic parallax); these can be mitigated using magnitude/phase corrections and direction remapping from head-centered to ear-centered coordinates. The second consists of near-field effects caused by head and torso geometry (e.g., shadowing, reflections, proximity effects) that cannot be addressed by simple analytic corrections. By distinguishing correctable head-centered reference effects from shape-driven near-field effects, the chapter provides guidance for both HRTF acquisition and use: it clarifies when head-centered normalization errors are perceptually negligible at typical measurement distances, when parallax-aware remapping is needed for near-field rendering, and why corrected far-field HRTFs can still deviate from true near-field HRTFs due to torso-related effects. Overall, this thesis presents and evaluates a hybrid room-acoustic prediction framework. Its development naturally motivated a closer examination of two core assumptions that affect practical use: the validity of single-decay statistical theory under imperfectly diffuse conditions, and the consequences of using far-field HRTFs in near-field binaural rendering. By clarifying when these assumptions are acceptable, when they fail, and how deviations map to perceptually relevant errors, the thesis provides practical guidance for extending hybrid room-acoustic simulations to more complex geometries, heterogeneous material distributions, and near-field interactive applications in both research and practice.