基于改进HHT的语音端点检测  

Speech Endpoint Detection Based on Improved HHT

作　　者：章小兵[1] 李燕萍[1] 王双杰[1] 

机构地区：[1]安徽工业大学电气与信息工程学院,安徽马鞍山243002

出　　处：《计算机工程》2016年第6期171-174,共4页Computer Engineering

基　　金：安徽工业大学产学研基金资助重大项目(RD14206003)

摘　　要：针对带噪语音在不同噪声环境下,尤其是非平稳噪声下难以判断语音段端点的问题,提出一种基于改进的希尔伯特-黄变换(HHT)瞬时能量的端点检测方法。对每帧带噪信号进行经验模式分解得到固有模态函数(IMF),选取分量IMF1,将其作为本底噪声并进行HHT得到噪声的瞬时能量,设置门限阈值,结合分量IMF3对带噪信号进行端点检测。该方法可提取非平稳噪声下带噪语音中的噪声成分,避免传统方法中选取前几帧信号作为噪声的局限性,同时利用分量IMF3进行端点检测达到滤波的效果。实验结果表明,该方法在不同噪声环境和低信噪比条件下提高了带噪语音端点检测的准确率。Aiming at the problem that speech endpoint is difficult to detect in different noise background,especially in non-stationary noise,an effective endpoint detection method is proposed based on Hilbert-Huang Transform（HHT） of instantaneous energy in the low Signal-to-noise Ratio（SNR） environment.Every frame of signal is decomposed into finite Intrinsic Mode Functions（IMF） by Empirical Mode Decomposition（EMD）,the instantaneous energy of IMF1 is calculated to get the energy value and combine with the IMF3 to detect the speech endpoint.The proposed method can effectively extract the noise component in the non-stationary noise,and avoid the limitation about the traditional method of selecting the former several frames as the noise,meanwhile selecting IMF3 to endpoint detection can achieve the effect of the filter.Experimental results show that this algorithm improves the endpoint detection accuracy in different noise background and the low SNR environment.

关 键 词：带噪语音 端点检测 希尔伯特-黄变换 瞬时能量 本底噪声 门限阈值 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]