Multichannel filter for enhancement of speech blocks

This work presents the concepts and the achieved results of a proposed microphone array algorithm based on multi-dimensional Wiener filter developed to work on blocks of speech. The inputs to the algorithm are two correlation matrices: the correlation matrix of the background noise affecting the desired signal and the correlation matrix of the signal affected by the noise. Experiments show that improvements of more than 12dB on signal to noise ratio can be achieved when comparing the filtered signals with one of the microphone array channels. In order to save computational load, the input signal is processed in blocks of a specified size and a technique is proposed to reduce blocking effects on the output filtered signal. It will be shown that practically there are no blocking effects. It is also shown that the technique is independent of the array physical configuration.