基于经验模态分解和Mel倒谱系数的语音端点检测  

Speech Endpoint Detection Based on the Empirical Mode Decomposition and Mel Cepstrum Coefficient

作　　者：陈蔚[1] 熊卫华[1] 施巍巍[1] 

机构地区：[1]浙江理工大学机械与自动控制学院,杭州310018

出　　处：《浙江理工大学学报（自然科学版）》2015年第4期574-578,共5页Journal of Zhejiang Sci-Tech University(Natural Sciences)

基　　金：浙江省自然科学基金项目(LY12F03011;LY12F03013)

摘　　要：提出了一种基于经验模态分解和Mel倒谱系数的语音端点检测方法。对语音信号进行分解得到一组IMF分量,将集中在低层IMF分量中的噪声信号滤除,重构剩余的IMF分量成语音信号,提取重构信号的Mel频率倒谱系数来检测语音信号的端点。实验结果表明,提出的方法可以较好地消除噪声对语音信号端点检测带来的影响,能在不同信噪比的环境下正确地对语音信号进行端点检测,具有一定的鲁棒性。This paper puts forward speech endpoint detection method based on empirical mode decomposition and Mel cepstrum coefficient.A group of IMF components were gained through decomposing speech signal.Noise signals concentrating at low-layer IMF components were filtered.The surplus IMF components were reconstructed to speech signal.Mel frequency cepstrum coefficient of reconstructed signal was extracted to detect endpoint of speech signal.The results show that this method can well eliminate effects of noise on endpoint detection of speech signal,and correctly detect endpoint of speech signal under different signal-to-noise ratios.This method has certain robustness.

关 键 词：语音端点检测 EMD IMF MFCC 信噪比 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]