【摘要】 为了建立一个数字串说话人身份确认系统 ,采用了基于数字HMM的与文本有关的身份确认方式 ;使用限定在数字范围内的任意数字串为语音文本 ,并内嵌能较好反映说话人特征的关键数字来提高确认性能 ;最终对多个关键数字的确认子系统进行综合判决 ,以提高HMM识别性能鲁棒性 .文中重点研究了基于关键数字HMM的身份确认 ,提出了以模型间的距离测度为准则来选择说话人的关键数字 ,并通过对比实验表明了用计算KLD上限的方法进行模型间距离测度的有效性 .更多还原

【Abstract】 In this paper, a speaker verification system of digit strings is proposed and the text dependent speaker verification based on digit HMM is adopted. By using the arbitrary digit string that includes key digits for the characteristics of the speaker, high performance is achieved. Then a comprchengive appraisal of several speaker verification subsystems based on key digits results in improvement of robustness. Through emphatically studying the speaker verification based on HMM of key digit, the method of choosing key digit ruled with distance measures between models is proposed. The experiment shows the validity of calculating distance measures between models with an upper bound for the KLD.更多还原