A Deep Learning-Based Pressure Matching Approach To Soundfield Synthesis

In this paper we propose a technique for soundfield synthesis based on the combination of the Pressure Matching (PM) approach and of deep learning-based methods. The pressure matching approach retrieves the driving signals for soundfield reproduction by minimizing the reproduction error at discrete control points through least squares. In this paper we follow a similar approach, but we perform the minimization by applying a Convolutional Neural Network (CNN). Through simulations, we compare the performance of the original pressure matching approach with the proposed technique and demonstrate how the latter is able to overcome spatial aliasing issues.