Abstract: The text-independent speaker recognition system based on Mel cepstral coefficients and GMM is discussed. In order to improve the recognition accuracy in noisy circumstance, antinoise methods are proposed in two aspects, one is based on speech recognition which changes text-independent system to text-dependent system, the other is based on frame selection method which has better robustness to make speaker recognition. The improvement effects to speaker recognition system of the two methods are analyzed, and it is proved by experiments that the recognition accuracy is enhanced in noisy circumstance by the methods given.

Key words: speech recognition, speaker recognition, text-independent, Mel cepstral coefficients, Gaussian mixed model(GMM)