Abstract: Independent Component Analysis(ICA) is a new method to separate the observed mixtures so as to estimate the independent components and can be used to extract the feature bases functions.The characteristic of the cepstrum vector agrees with the assumption of the ICA transformation.ICA bases is transformed from improved robust Mel Frequency Cepstrum Coefficients(MFCC) by ICA and then used in speaker recognition.A novel speaker recognition system based on ICA is presented.The results of experiments show that the proposed system has a higher recognition rate in noisy environments.