文本无关的说话人识别系统抗噪方法研究  

Research on Antinoise Methods of Text-independent Speaker Recognition System

作　　者：叶蕾[1] 方鹏[2] 

机构地区：[1]南京邮电大学通信与信息工程学院,江苏南京210003 [2]中国移动通信集团江苏有限公司,江苏南京210029

出　　处：《计算机与现代化》2010年第10期16-19,共4页Computer and Modernization

摘　　要：研究了基于美尔倒谱特征参数及高斯混合模型的文本无关的说话人识别系统,为了提高噪声环境下识别系统的识别率,从两个角度研究改善该系统抗噪性能的方法,即利用语音识别将文本无关的系统转化为文本有关的说话人识别方法和通过选择鲁棒性较强的帧进行说话人识别的方法,分析了以上方法对系统识别性能的改善作用,并通过实验验证上述方法确实可以提高系统在噪声环境下的识别率。The text-independent speaker recognition system based on Mel cepstral coefficients and GMM is discussed.In order to improve the recognition accuracy in noisy circumstance,antinoise methods are proposed in two aspects,one is based on speech recognition which changes text-independent system to text-dependent system,the other is based on frame selection method which has better robustness to make speaker recognition.The improvement effects to speaker recognition system of the two methods are analyzed,and it is proved by experiments that the recognition accuracy is enhanced in noisy circumstance by the methods given.

关 键 词：语音识别 说话人识别 文本无关 美尔倒谱参数 高斯混合模型 

分 类 号：TN912.3[电子电信—通信与信息系统]