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

Area of Science:

• Acoustics
• Computational physics
• Machine learning

Background:

• Accurate acoustic simulations are crucial for virtual/augmented reality, game audio, and spatial computing.
• Traditional numerical methods for solving the wave equation are computationally intensive, especially for parametric source positions.
• This limits the practical application of detailed acoustic simulations in dynamic virtual environments.

Purpose of the Study:

• To develop a computationally efficient method for predicting sound propagation in 3D virtual rooms with parametric source positions.
• To overcome the limitations of conventional numerical discretization methods for wave phenomena.
• To enable real-time acoustic simulations for applications like virtual reality and game audio.

Main Methods:

• Utilizing deep operator networks to approximate linear wave-equation operators.
• Developing a compact surrogate model to avoid pre-calculating and storing impulse responses.
• Applying the method to various complex 3D acoustic scene geometries.

Main Results:

• Achieved millisecond-scale computation times for sound propagation prediction.
• Demonstrated good agreement with reference solutions across different scene complexities.
• Reported root mean squared errors between 0.02 and 0.10 Pa.

Conclusions:

• Deep operator networks offer a computationally efficient solution for 3D acoustic simulations with parametric source positions.
• The proposed method represents a paradigm shift, enabling precise wave field predictions in realistic domains.
• This advancement facilitates more immersive and realistic audio experiences in virtual and augmented reality applications.