Efficient modeling of sound scattering in rooms
Dingding Xie defended her PhD thesis at the Department of Built Environment on April 30.
The growing demand for perceptually credible and computationally efficient room acoustic simulations has driven advances in acoustic modeling, particularly for applications such as auralization and Acoustic Virtual Reality. A key challenge lies in accurately representing sound scattering caused by surface roughness, room geometry, and interior objects, without incurring the high computational costs of wave-based methods.
In her research, developed compact, physically meaningful, and perceptually relevant scattering descriptions that could be integrated into hybrid room acoustic models, balancing accuracy and efficiency.
Average and directional scattering characterization
The thesis introduced frequency-dependent Average Scattering Coefficients (ASCs) as a room-averaged measure of scattering derived from coherence analyses of impulse responses in furnished and empty rooms. The results showed that ASCs mainly depended on the amount of interior elements and absorption, were largely independent of source鈥搑eceiver configurations, and effectively controlled the transition from coherent to incoherent sound energy in hybrid models.
Building on this work, Directional Coherence Loss Coefficients (DCLCs) were proposed to describe the angular distribution of scattering-induced coherence loss from a receiver鈥檚 perspective. These coefficients captured directional and localized scattering effects that could not be represented by scalar measures alone.
Hybrid sound-field reconstruction and validation
ASCs and DCLCs were incorporated into hybrid sound-field reconstruction frameworks combining coherent specular components with stochastically modeled incoherent components. Validation studies demonstrated good agreement with wave-based simulations and measurements in terms of energy decay, reverberation time, clarity, diffuseness, and spatial correlation, confirming that aggregated scattering could be modeled without explicitly representing detailed geometries.
Extended models and data-driven estimation
Xie's thesis further developed a hybrid model combining the Image Source Method and Acoustic Radiosity while preserving phase information, enabling the simulation of directional, non-diffuse sound fields in rooms with non-uniform absorption or scattering.
To address the challenge of obtaining surface scattering coefficients, an inverse relationship between DCLCs and surface scattering coefficients was established. A neural network trained on simulated data successfully estimated scattering coefficients with errors below perceptual just-noticeable differences.
Overall, Xie's dissertation presented a coherent framework for efficient and perceptually relevant modeling of room sound scattering. By integrating ASCs and DCLCs into hybrid models, extending Acoustic Radiosity, and introducing a data-driven estimation approach, the work advanced state-of-the-art room acoustic simulation techniques.
Title of PhD thesis: Supervisors: Maarten Hornikx and Marco Berzborn