What is it about?

We propose a method capable of accurately predicting the acoustics in the millisecond range for any combinations of sound source and listener positions in virtual rooms and buildings. With previous numerical methods, this quickly got intractable due to computational challenges. We have used a novel scientific machine-learning approach to predict 3D wave propagation influenced by the source position, room geometry, and damping wall materials.

Featured Image

Why is it important?

Modeling realistic acoustics in virtual environments are important for immersive experiences, e.g., the experience of being in a church is very different than being in a living room. Our method learns the underlying physics, yet predicting in the millisecond range crucial for real-time virtual environment scenarios. We demonstrated our model in four geometries of increased complexity, including a dome with complex geometries, showing its capability of learning complex physical behavior.


I am very excited about the increased focus on realistic acoustics with the introduction of spatial computing and the rapid introduction of new VR and AR headsets. Historically, graphics in computer games have seen higher priorities than audio rendering -- but not anymore. With new immersive experiences blurring the lines between the real and the virtual worlds, we need new methods for simulating accurate real-time acoustics. I hope this contribution will be inspiring.

Nikolas Borrel-Jensen
Technical University of Denmark

Read the Original

This page is a summary of: Sound propagation in realistic interactive 3D scenes with parameterized sources using deep neural operators, Proceedings of the National Academy of Sciences, January 2024, Proceedings of the National Academy of Sciences,
DOI: 10.1073/pnas.2312159120.
You can read the full text:




The following have contributed to this page