基于矢量量化(VQ)的说话人识别的研究-计算机技术专业论文.docxVIP

研究生优秀毕业论文 Abstract Abstract 硕士学位论文 Abstract Speal(er recognition tecllIlology is an important research branch in speech recognition field．Its identification is based on the feature parameters，which are extracted to ea’ectively renect the personality of people．Speaker recognition process includes pre-processing of the speech signal，feature extraction，modeling呻d model matching．This p印er made the following research： In tenlls of speech enhancement，due to the influence of noise on the perfbnnance of speaker recognition system，FaStICA based on the negatiVe entropy was 11ighlighted．In this paper，FastICA was combined with STSA．MMSE a11d was used in仔ont．end speech enhancement．Experimentalresults show that the e肫ct of speech e1111aJlcement is obvious． In endpoint detection stage，the paper studied the traditional endpoint detection、矿hich is based on dual threshold and cepstmm distance．Bo玎(}谢ng idea from t11is method，aJl inlproVed cepstnⅡn distance endpoint detection algorithm was proposed．The comparative experiments show that tlle improved method can achieve better results． In f．eature extraction stage，the cepstral fIeanJre and pitch of speech signal were combined as feature parameters in speaker recognition system．However，if mese feature p猢eters were superimposed directly，me锄ount of calculation、析ll increases，thereby both the n．aining and the recognition time will also increase．So Fisher critedon、vas used to select feature dimensions． First，the Fisher criterion ratio con．esponding to each dimension of feature parameters was calculated．Then，seveml dimensions of each feature， which correspond to seVeral leading biggest Fisher criterion ratios，was selected．Among these combined features，one group which can achieve the best recognition pe怕mlaJlce was adopted．Experimental results show tha