Sound propagation in realistic interactive 3D scenes with parameterized sources using deep neural operators

Significance Solving partial differential equations is computationally expensive, creating challenges for real-time physics simulations involving the wave equation in virtual acoustics—e.g., mixed reality, spatial computing, and metaverses. Considering dynamic scenes with many source and receiver positions further increases the computational effort required. To overcome this, we propose using deep neural operators along with domain decomposition and transfer learning frameworks, to predict pressure fields accurately and efficiently in real-time for complex three-dimensional environments. In an example of a dome measuring 36 m3, with intricate geometries, we showcase effective prediction of full-wave propagation regardless of sound source and receiver positions, on a scale and accuracy not demonstrated earlier, hence paving the way for unprecedented possibilities in future immersive experiences.