A multi‐channel subband generalized singular value decomposition approach to speech enhancement

Abstract In speech communication applications, the presence of noise (and reverberation) in the environment may render the captured speech unintelligible. Noise reduction algorithms are thus needed to improve the intelligibility. Recently, a multi‐channel noise reduction technique, based on a Generalized Singular Value Decomposition (GSVD) has been proposed. The GSVD based filter minimizes the Mean Square Error (MSE) between the desired signal portion in the received signals and the sum of the filtered received microphone signals. In this paper, we propose a subband implementation of the GSVD based filter. It is shown that — in the case of coloured signals — the subband implementation improves intelligibility more than the fullband approach. In addition, a significant complexity reduction is achieved, allowing for low cost, real‐time applications.