A New Method for Blind Source Extraction

Tsubasa YOSHIHARA, Kiyotoshi MATSUOKA
2009 SICE Journal of Control Measurement and System Integration  
Blind source separation (BSS) is a method for recovering a set of statistically independent signals from the observation of their mixtures without any prior knowledge about the mixing process. If, as a special case, only one source component is to be extracted, it is called blind source extraction (BSE). Since BSE involves a smaller number of parameters to be estimated than BSS, it requires less computation time. In this paper we propose a new algorithm for BSE of the convolutive mixture. The
more » ... tive mixture. The algorithm determines the extractor by evaluating independence between a target signal and other signals. It has some good properties. First, since it is formulated in the time domain, we do not need to worry about the so-called permutation problem. Second, by applying a particular constraint on the extractor the signal quality at the sensors is preserved through the extraction process. A couple of experiments are shown in which the proposed algorithm is applied to the mixture of five voice signals.
doi:10.9746/jcmsi.2.192 fatcat:2i5kwwgdw5ehhee3x57q3syeoq