文摘
Past research has proven that a first-order B-format ambisonic signal can be used to partially reconstruct the original sound field through a collection of arbitrary positioned loudspeakers. This is achieved by setting the gain of each loudspeaker to be the weighted sum of the three components in the B-format signal. Deduction of the weighting factors (a.k.a. the decoding parameters) has been successfully accomplished with the use of the Modified Tabu Search (MTS), and later with the Heuristic Genetic Algorithm (HGA) which provides higher precision and stability. Despite the favorable outcome, both methods involve large amount of iterations and the computation time is lengthy. In this paper, we propose a scheme to overcome this problem based on the integration of Neural Network Estimation (NNE) and the MTS. Compared to HGA, the new approach is about two orders of magnitude faster, and at the same time capable of attaining similar precision in determining the decoding parameters.