Abstarct: With the increasing advance in science and technology, the human needs in all areas of intelligent systems have become more obvious. As a person’s voice like every other identifiers such as fingerprint, facial features, iris, etc is unique, it can be used in speaker recognition smart systems. But one of the most important issues in the field of speaker recognition systems is the effect of noise on the speech signal that may lead to a decrease in the recognition accuracy. Due to the importance of robust speaker recognition in noise (or noisy conditions), many studies recently investigated the issue and various methods have been proposed. The purpose of this thesis is to use the time-frequency distribution of Wigner-Ville for feature extraction. The study employed the combination of Wigner-Ville distribution function and Hillbert transform and MFCC coefficient for feature extraction. In this method, after pre-emphasis and using window, Hilbert transform of speech signal is obtained and then the signal is analysed by Wigner-Ville transformation .The output signal obtained from Wigner-Ville transform is passed through the Mel Filter Bank and after taking the logarithm, the cosine Fourier transform is used. The output of the proposed system is used as a feature vector for speaker recognition. A GMM model also is used to model any speaker. The results of proposed method are compared with MFCC and PLP coefficients. The experiment indicates that the proposed method provides a better results than MFCC and PLP coefficients in low signal to noise ratio.