Abstract: With the purpose of improving the performance of speaker recognition, an adaptive speaker recognition method based on GMM is proposed. It can automatically select different length of speech for different speakers so as to reduce the recognition time through two aspects: speaker acoustic features calculation and recognition probability estimation. So it can remarkably improve the recognition speed than customary methods while keeping the correct recognition ratio. Experiments show that the recognition speed is increased about 4 times while keeping the recognition ratio at the level of 97%. This novel method is very fit for large muster of speaker recognition based on GMM.


Key words: speaker recognition, Gaussian mixture model (GMM), linear prediction coefficient (LPC), adaptation