Data-driven simulation for two-dimentional sound field considering room shape

Estimating the sound fields in a room using numerical simulations based on the wave equation typically requires extensive computation time. Consequently, several recent studies have explored the use of deep learning to reduce computation time. However, these studies have not adequately accounted for differences in room shapes, which significantly affect reverberation. In this study, we propose a novel approach that utilizes sound field simulation results for deep learning training data to estimate the sound field in rooms of various shapes. Our target data were obtained using the boundary element method, and we employed full-resolution residual networks as our deep learning models. Through our experiments, we evaluated the accuracy of our proposed method for sound field estimation as well as the computational time required.