端点检测对说话人识别的影响  

Impact of Endpoint Detection on Seaker Recognition

作　　者：顾秀秀[1] 赵玉晓[1] 张二华[1] 

机构地区：[1]南京理工大学计算机科学与技术学院,南京210094

出　　处：《计算机与数字工程》2014年第6期984-987,共4页Computer & Digital Engineering

摘　　要：由于在说话人识别研究中发现,语音信号包括静音段、辅音段和浊音段,说话人的个性特征主要蕴含在浊音段中,静音段与辅音段参与识别会明显降低说话人的识别率。同时大量的实验证明,使用端点检测去掉静音段和辅音段后识别率有明显的提升,针对这种情况,指出了端点检测后的语音,其特征分布将会更加符合高斯分布。实验结果表明,端点检测越精确,其识别效果也就越好。Since the discovery in speaker recognition research ,the voice signal includes mute segment ,consonants seg-ment and voiced segment in which the main speaker＇s personality traits inherent .Mute segment and consonant segment in-volved in recognition will significantly reduce the rate of speaker recognition .A number of experiments demonstrate that re-moving of mute segment and consonants segment with endpoint detection has significantly improved the recognition rate .For this situation ,the speech after endpoint detection is pointed out ,whose distribution will be more in line with a Gaussian dis-tribution .The experimental results show that the more accurate endpoint detection ,the recognition result is better .

关 键 词：端点检测 特征提取 短时能量 短时过零率 MEL频率倒谱参数 

分 类 号：TN912.34[电子电信—通信与信息系统]