Abstract
An integrated recording and reproduction array system for spatial audio is presented. The system has a generic framework akin to the analysis-synthesis filterbanks in discrete time signal processing. The architecture of the system consists of two stages. In the first stage, an analysis microphone array that “encodes” the sound field into sparse source components in the angle space. Direction of Arrival (DOA) of sound sources is estimated by MUltiple SIgnal Classification (MUSIC). The second stage is a field synthesis by loudspeaker array that “decodes”, or reproduces, the sound field on the basis of the source components obtained in the first stage. Specifically, this stage can be formulated as a deconvolution problem by matching the reproduced and target fields at numerous control points in the interior domain of the loudspeaker array, while minimizing the external radiation to mitigate the adverse effects of boundary reflections. The inverse problem is solved by Truncated Singular Value Decomposition (TSVD) or Convex Optimization (CVX) algorithms. Since the performance of sound field synthesis may be constrained by aliasing frequency, a hybrid method combing deconvolution and vector panning is proposed here for improving the reproduction. The result of source extraction and the performance of reproduction are also evaluated.