Physics-informed neural networks for one-dimensional sound field predictions with parameterized sources and impedance boundaries
This work addresses the problem of handling dynamic acoustics in virtual environments like games and mixed reality, though it is incremental as a first step toward 3D applications.
The paper tackled the challenge of efficiently predicting sound fields in dynamic scenes with moving sources by developing a physics-informed neural network (PINN) method for 1D scenarios, achieving relative mean errors below 2% or 0.2 dB.
Realistic sound is essential in virtual environments, such as computer games and mixed reality. Efficient and accurate numerical methods for pre-calculating acoustics have been developed over the last decade; however, pre-calculating acoustics makes handling dynamic scenes with moving sources challenging, requiring intractable memory storage. A physics-informed neural network (PINN) method in 1D is presented, which learns a compact and efficient surrogate model with parameterized moving Gaussian sources and impedance boundaries, and satisfies a system of coupled equations. The model shows relative mean errors below 2%/0.2 dB and proposes a first step in developing PINNs for realistic 3D scenes.