说话人识别的不确定性i-vector分析  

Uncertainty modified i-vector for speaker recognition

作　　者：屈召贵 鲁顺昌 

机构地区：[1]四川工商学院计算机系,四川成都611745

出　　处：《计算机工程与设计》2017年第6期1647-1650,共4页Computer Engineering and Design

基　　金：四川省教育厅科研基金项目(16ZA0424);四川工商学院创新科研基金项目(2013cscdky0046)

摘　　要：针对噪声环境中说话人识别性能不稳定问题,提出一种基于不确定性前端因子分析的说话人识别方法。通过不确定性估计改进传统的i-vector特征抽取方式,实现在噪声环境中性能稳定的说话人识别。实验结果表明,该方法具有较高的说话人识别准确率,是高鲁棒性的,可广泛用于语音识别任务。An uncertainty front end factor analysis based speaker recognition method was proposed focusing on the problem of noise-sensitivity.Traditional i-vector extraction was modified according to uncertainty estimation and reliable performance was achieved in noisy environment.Experimental results reveal the presented method has higher accuracy,which is more robust than many other algorithms.It is an outstanding tool in audio processing related fields.

关 键 词：说话人识别 不确定性 鲁棒性 i-vector 前端因子分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]