Based Text-independent Speaker Identification Technology

This paper starts around the construction of text-independent speaker recognition system. The pretreatment of voice signal,feature extraction,establishment of speaker model and judgement strategy are mainly studied. Two speaker recognition systems have been established: one is based on VQ and the other is GMM. There are several aspects can be shown in the following:At the aspect of voice signal pretreatment, Firstly, the sampling rate is set to be 8kHz and the quantization rate is 16bit. Secondly, the speech signal is treated with pre-emphasis and windowing frame with Hamming.At the aspect of feature extraction, Some feature parameters such as LPC,LPCC and MFCC are extracted. Their performance under GMM shows that MFCC is the best, then is LPCC and the worst is LPC.At the aspect of model recognition, VQ, HMM and GMM are introduced in this paper. Through the test of speaker identification and speaker verification, it shows that GMM is better than VQ.Under the system of GMM, On the same test condition, the impact of the order of GMM on system recognition rate is studied. The negative impact of higher or lower order is analyzed, and the choice is made according to the practical circumstance; Setting covariance threshold in EM algorithm iteration is put forward. The comparison of experiments on different threshold is made, finding 0.10 is a universal and practical value for the covariance threshold; Putting forward a combination of divisive method and k-means clustering method on the basis of general method on initializing the GMM's parameters, the result shows that the improved method can promote the recognition rate.In the end, a conclusion of this paper and prospect of the future work are drawn.