Extracting a source of shorter source-to-microphone distance from convolutive mixtures

Abstract

Although blind source extraction is advantageous over blind signal separation in practical applications, it is difficult to choose the extracted source from many sources in the mixtures. A new algorithm is presented to extract a source signal of shorter source-to-microphone distance from two convolutive mixtures. The developed algorithm imposes an additional constraint on the extracting filters, i.e. high direct-to-reverberation ratio. In experiments, the algorithm successfully extracted the shorter-distance signal even though the interference was much stronger for both simulated convolutive mixtures and real-recorded mixtures in an office room.