Speaker
Description
Diffusion is a key, but often underweighted characteristic of acoustic surface design in interior spaces. While absorption and reflection dominate practical acoustic treatment, diffusion governs the uniformity of sound propagation, particularly in larger auditoriums and multipurpose halls. Algorithmically generated diffusive geometries deliver calculable scattering performance but are often visually repetitive and aesthetically constrained, limiting their integration into architecturally expressive interiors. This study investigates whether image-based generative AI can produce visually diverse diffusion panels that remain acoustically viable compared with algorithmic baselines.Two-generation strategies are compared. The first is an algorithmic pipeline producing diffuser geometries tuned to a defined target frequency band. The second is an AI-driven pipeline that combines image generation with language-model-assisted prompting for aesthetic variation, followed by translation into surface geometry suitable for acoustic evaluation. Candidate panels from both pipelines are filtered using geometric heuristics related to depth variation, spatial distribution, and feature scale before acoustic analysis.Selected panels are evaluated through simulation at both panel and room scale, building on prior cross-validation work by the authors. The comparison aims to quantify the acoustic cost of aesthetic freedom and to identify conditions under which AI-generated diffusers remain within an acceptable margin of algorithmically generated optima. The work sits at the intersection of architectural acoustics, parametric design, and generative AI, extending a previously published generative pipeline by specializing it for acoustic diffusion and introducing quantitative acoustic validation as a selection criterion.