Mathematical metric of speech distortion in various types of BSSA

摘要

In this paper, we propose a structure-generalized parametric blind spatial subtraction array (BSSA), and the theoretical analysis of the amounts of musical noise and speech distortion is conducted via higher-order statistics. We theoretically prove a tradeoff between the amounts of musical noise and speech distortion in various BSSA structures. From the analysis and experimental evaluations, we reveal that the structure should be carefully selected according to the application, i.e., a channelwise BSSA structure is recommended for listening but a normal BSSA is more suitable for speech recognition.