基于希尔伯特-黄变换的低信噪比语音端点检测  

Voice activity detection with low signal-to-noise ratio based on Hilbert-Huang transform

作　　者：刘柏森[1,2] 卢志茂 申丽然 金辉[1] 

机构地区：[1]哈尔滨工程大学信息与通信工程学院,哈尔滨150001 [2]黑龙江工程学院电子工程系,哈尔滨150050

出　　处：《吉林大学学报（工学版）》2011年第3期844-848,共5页Journal of Jilin University:Engineering and Technology Edition

基　　金：国家自然科学基金项目(60603092;60803087);高等学校博士学科点专项科研基金项目(20070217043)

摘　　要：应用希尔伯特-黄变换完成了一种低信噪比条件下的语言信号端点检测。该方法通过分析纯净语音信号与低信噪比下语音信号的固有模态函数及希尔伯特谱,找出固有模态函数中语音信号能量集中的分量,分析其希尔伯特谱,自适应地选取阈值进行语音段与非语音段的检测。通过对比实验表明了该方法在低信噪比下能有效地检测出语音信号。Voice activity detection plays an important role in speech signal processing. However, commonly used detection methods are not valid in low Signal-to-Noise Ratios （SNR）, and the voice activity detection becomes difficult. Hilbert-Huang Transform （HHT） is a method designed for non- linear and non-stationary signals. Because of its adaptability, HHT can be used for analysis of available data. In this paper, the HHT is applied for voice activity detection under the condition of low SNR. This method finds the concentration component of the Intrinsic Mode Function （IMF） in the speech signal energy, analyzes the Hilbert spectrum by using the IMF and the Hilbert spectrum between pure voice signal and voice signal under low SNR. It can adaptively select threshold for voice and non-voice segment detection. Comparison of experiment results shows that this method is effective under low SNR of the detected speech signal.

关 键 词：信息处理技术 希尔伯特-黄变换 经验模态分解 语音检测 

分 类 号：TN912.3[电子电信—通信与信息系统]