Abstract:  Blind signal extraction is a hot research field of blind signal processing，it can only extract the interesting signals, reduce the computing consumption effectively, and overcome the difficult problem is that the sequence uncertainty of the separated signals of which the simultaneous separation of blind source separation, it plays an important role in many areas such as biomedical signal analysis and processing (EEG: electroencephalogram, MEG: magnetoencephalography, fMRI: functional magnetic resonance imaging), speech and image recognition and enhancement, and so on. The drawback of blind source extraction algorithm based on temporally structure is that too sensitive to the estimation error of the time delay and has weak noise immunity, this paper combining skewness and temporal structure as a new cost function to conquer that defects. First of all, this algorithm uses the characteristic of asymmetry on skewness to evaluate the nonGaussian to suppress the noise, and when run the algorithm program on computer, compared with the traditional kurtosis measure method this algorithm reduces the computing consumption at the same time, in the next place paper uses the pitch of the expect signal as the best time delay estimation to successful extract the interesting source. When joining the skewness and temporal structure, the algorithm is insensitive to the estimation error and is robust for the noise. After that this paper gives specific operational steps of this algorithm. In the blind signal extraction computer simulation and experiments part, three speech signals is selected from the standard TIMIT corpus are one man and two women reading the same statement separately, the blind signal extraction experimental results indicating that compared with the literature 3, this algorithm has better separation efficiency and have faster computing speed, compared with the literature 4 both the algorithms have better separation efficiency but the algorithm extraction rate of this paper is more quickly, which is confirm the effectiveness of this algorithm.