基于VEMAP的说话人识别鲁棒性研究  

Research on robustness of speaker recognition based on VEMAP

作　　者：黄文娜[1] 彭亚雄[1] 贺松[1] HUANG Wenna PENG Yaxiong HE Song(College of Big Data and Information Engineering, Guizhou University, Guiyang 550025, Chin)

机构地区：[1]贵州大学大数据与信息工程学院,贵州贵阳550025

出　　处：《电声技术》2016年第11期44-47,共4页Audio Engineering

摘　　要：为了改善发声力度变化对说话人识别系统性能的影响。针对不同发声力度下语音信号的分析,提出了使用发声力度最大后验概率(Vocal Effort Maximum A Posteriori,VEMAP)自适应方法更新基于高斯混合模型-通用背景模型(Gaussian Mixture Model-Universal Background Model,GMM-UBM)的说话人识别系统模型。实验表明,所提出的方法使不同发声力度下系统EER%降低了88.45%与85.16%,有效解决了因发声力度变化引起的训练语音与测试语音音量失配,从而导致说话人识别性能降低的问题,改善说话人识别系统性能效果显著。In order to improve the performance of recognition system caused which is influenced by the changes of vocal ef- forts. In this paper, based on the analysis of the speech signals under different vocal efforts, Vocal Effort Maximum A Poste- riori （VEMAP） adaptive method is proposed to update the speaker recognition model which based on Gaussian Mixture Mod- el-Universal Background Model （GMM-UBM）. From the results of this experiment, VEMAP adaptive method can make the EER% of recognition system which under different vocal efforts reduced by 88.45% and 85. 16% , which effectively solve the vocal mismatch of training speech and test speech caused by vocal effort that lead to the speaker recognition performance degradation problem, improving the speaker recognition system performance obviously.

关 键 词：说话人识别 发声力度 发声力度最大后验概率自适应 高斯混合模型-通用背景模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]