Online target-speech extraction method for robust automatic speech recognition

The present invention relates to a method of separating a target voice signal for robust voice recognition. The method comprise the following steps: (a) receiving information on direction of arrival of target sound source for a microphone; (b) eliminating the target voice signal from input signals and creating a nullformer for estimating a noise by using the information on the direction of arrival of the target sound source; (C) setting the target-sound source to a first channel, and a dummy output by the nullformer to remaining channels by using an adaptive form vector(W(K)); (d) setting a cost function for minimizing dependency between a real output of the target sound source and the dummy output using the nullfirmer by using an independent component analysis; (e) estimating the target sound source by using the cost function; and whereby the target voice signal is separated from the input signals.