摘要： 针对声纹密码识别中声纹文本信息利用率低和噪音干扰的问题,提出隐马尔科夫模型-通用背景模型(HMM-UBM)融合相关向量机（RVM）的声纹识别算法。利用HMM-UBM对语音信号进行时序建模,使用RVM学习得到每位注册话者语音的分类信息。通过对待识别话者建立HMM模型，并将RVM作为分类器进行判决决策得到分类结果。实验结果表明,与GMM-UBM算法和GMM-SVM算法相比,在无噪声环境下,该算法错误接收概率降低7%~9%,识别正确率提高4%~5%，在低信噪比环境下,其识别正确率提高5%~12%。

关键词: 声纹识别, 隐马尔科夫模型, 通用背景模型, 相关向量机, 语音信号

Abstract: Considering shortcomings of noise interference and low utilization in voiceprint recognition,a voiceprint recognition algorithm which is based on Hidden Markov Model-Universal Background Model(HMM-UBM) and Relevant Vector Machine(RVM) is proposed.The algorithm uses HMM-UBM to model speech signals in time sequence,and learn the classified information of each registered speaker by using RVM.It is established the HMM model for the speaker who needs to be identified,and the results of classification is obtained by using RVM as classifier.Experimental results show that compared with the GMM-UBM algorithm and GMM-SVM algorithm in a noiseless environment,the probability of error acceptance of proposed algorithm is reduced by 7%~9% and the accuracy of recognition is improved by 4%~5%.In low SNRR environment,the accuracy of recognition is improved by 5%~12%.

Key words: voiceprint recognition, Hidden Markov Model(HMM), Universal Background Model(UBM), Relevance Vector Machine(RVM), speech signal

中图分类号: 

• TP309