基于谱减法及频谱方差的语音端点检测方法  

A Voice Activity Detection Algorithm Based on Spectral Subtraction and Spectrum Variance

作　　者：刘新飞 周辉 

机构地区：[1]装备学院信息装备系,北京101416

出　　处：《兵工自动化》2014年第4期34-38,共5页Ordnance Industry Automation

摘　　要：为提高在强背景噪声条件下语音端点检测的性能,提出一种将改进的谱减法和频谱方差相结合的方法。先利用谱减法对带噪语音信号进行增强处理,然后使用基于频谱方差的方法进行检测,并以一男一女对10个数字和26个英文字母各发音2次的录音作为样本进行仿真验证。实验结果表明:该方法计算简单,可靠性高,能有效减少背景噪声,提高语音信号的信噪比,在较低的信噪比下仍能比较准确地检测到语言信号的端点。For improving performance of voice activity detection （VAD） under big background noise, proposes an improved method by combining modified spectral subtraction and spectral variance. The proposed scheme employs modified spectral subtraction to enhance the speech firstly, and then detect based on spectral variance. And at the same time, a large number of simulation experiments have been done to confirm the validity of the scheme which take the pronunciations of 10 digits and 26 letters recorded each one twice by a man and an woman as testing samples. Experiment result shows that the method simple calculation and high reliability, it can reduce background noise effectively, increase signal-to-noise ratio of voice signal, can accurately detection voice signal endpoint under low signal-to-noise ratio.

关 键 词：谱减法 频谱方差 端点检测 

分 类 号：TP216[自动化与计算机技术—检测技术与自动化装置]