摘要
为了提高语音激活检测在低信噪比环境中的检测性能,提出了一种基于奇异谱的语音激活检测方法。首先用多窗口方法计算每一帧语音信号的相关矩阵;然后对相关矩阵进行奇异值分解;利用奇异值可以反映有用信号和噪声分布情况的特性,将每一帧语音信号经过加权处理后的最大奇异值与自适应阈值进行比较进行语音激活检测。该方法原理简单,易于硬件实现,通过实验仿真表明,在低信噪比环境下,和基于对数能量方法相比,本文方法也能够很好的区分语音段和非语音段,有良好的检测性能。
In order to improve the performance of voice activation detection at low SNR(Signal to Noise Ratio), we proposed a detection approach of voice activity based on singular spectrum. Firstly, we calculate the correlation matrix for each frame of speech signal with multi-window approach; then performed singular value decomposition to the correlation matrix; due to the singular value reflects the characteristics of the useful signal and noise distribution, we can perform activity detection through comparing the weighted maximum singular value of each frame of speech signal with the adaptive threshold value. This method is simple and can be easily implemented in hardware. The simulation indicates that compared with energy method based on logarithm, in low SNR environment, this approach can better distinguish speech segments with non-voice segment better.
出处
《应用声学》
CSCD
北大核心
2013年第2期137-143,共7页
Journal of Applied Acoustics
基金
国家自然科学基金项目(61071196
61102131)
教育部新世纪优秀人才支持计划项目(NCET-10-0927)
信号与信息处理重庆市市级重点实验室建设项目(CSTC2009CA2003)
重庆市杰出青年基金项目(CSTC2011jjjq40002)
重庆市自然科学基金项目(CSTC2009BB2287
CSTC2010BB2398
CSTC2010BB2409
CSTC2010BB2411)资助
关键词
语音激活检测
Slepian数据窗
离散扁椭圆序列
相关矩阵
奇异值分解
自适应阈值
Voice activity detection, Slepian data window, Discrete prolate spheroidal sequences,Correlation matrix, Singular value decomposition, Adaptive threshold