Abstract
Circular microphone arrays (CMA) are preferred over more complex spherical microphone arrays (SMA) in the context of some audio applications because azimuthal angles of spatial sound are considered more important than the elevation angles in those scenarios. However, the fact that CMA does not resolve the elevation angle well can be a limitation for some applications which involves 3-dimensional sound fields. But this can also be a limitation in spatial audio rendering. Sources with elevation less than 60 degrees can be localized precisely respectively. This paper proposes a 2.5-dimensional (2.5-D) CMA that consists of an unbaffled CMA and a vertical logarithmic-spacing linear array on the top. In the localization stage, two delay-and-sum (DAS) beamformers are applied to the circular array and the linear array, respectively. The product of the identified angular patterns yields the direction of arrival (DOA). In the separation stage, Tikhonov Regularization (TIKR) and Compressive Sensing (CS) are employed to extract the source signal amplitudes from the output signals from two arrays. The extracted signals are further processed by Normalized Least-Mean-Square (NLMS) algorithm with Internal Iteration (IIT) Algorithm respectively in order to produce the source signal with improved quality. To validate the 2.5-D CMA experimentally, a three-dimensionally printed circular array comprised of a 24 micro-electro-mechanical-system (MEMS) microphone circular array and an 8- MEMS microphone logarithmic-spacing linear array is constructed for localization and separation for sound sources. Objective Perceptual Evaluation of Speech Quality (PESQ) test and a subjective listening test are undertaken in performance evaluation. The experimental results demonstrate better separation quality achieved by the CS combined with NLMS method than by the TIKR combined with NLMS method.