Implicit Neural Representation With Physics-Informed Neural Networks For The Reconstruction Of The Early Part Of Room Impulse Responses

Abstract

Recently, deep learning and machine learning approaches have been widely employed for various applications in acoustics. Nonetheless, in the area of sound field processing and reconstruction, classic methods based on the solutions of the wave equation are still widespread. Lately, physics-informed neural networks have been proposed as a deep learning paradigm for solving partial differential equations that govern physical phenomena, bridging the gap between purely data-driven and model-based methods. In this study, we exploit physics-informed neural networks to reconstruct the early part of missing room impulse responses in a uniform linear array. This methodology allows us to leverage the underlying law of acoustics, i.e., the wave equation, forcing the neural network to generate physically meaningful solutions given only a limited number of data points. The results from real measurements show that the proposed model achieves accurate reconstruction and performance in line with state-of-the-art deep learning and compressive sensing techniques while maintaining a lightweight architecture.