Abstract

Beyond the Frequency Divide: Rethinking Room Acoustic Simulation with Rays, Waves and AI

Room acoustic simulation has traditionally been taught as a map of separate territories: geometrical methods at high frequencies, wave methods at low frequencies, and a transition region that neither serves well. Yet that gap was never a failure of physics but a compromise between computational cost and acceptable error, drawn with the hardware and material data of its time.
We examine how advances in wave propagation, numerical computing, material modelling and machine learning are redrawing these frontiers. Rays now carry phase, diffraction and coherent field reconstruction into frequency ranges once reserved for wave methods. Time-domain wave solvers are becoming practical at room scale. Impedance-based boundary descriptions, including non-locally reacting materials, reveal how much of the disagreement between methods lies in the walls rather than the solvers. Physics-guided learned models, from modal surrogates to neural operators, change how often an expensive calculation needs to be repeated at all.

Drawing on the development of soundy.ai, we show how traditional and AI-based engines can operate within one platform, supporting comparison, validation and complementary use. What, then, should replace the frequency divide? We argue for an evidence-based answer: for each approach, the rooms, frequency bands and boundary assumptions under which it can be trusted.

About Csaba Huszty