基于多特征融合与动态阈值的语音端点检测方法  

Speech Endpoint Detection Method Based on Multi-Feature Fusion and Dynamic Threshold

作　　者：朱春利[1] 李昕[1] ZHU Chunli;LI Xin(School of Mechatronics Engineering and Automation,Shanghai University,Shanghai 200072,China)

机构地区：[1]上海大学机电工程与自动化学院,上海200072

出　　处：《计算机工程》2019年第2期250-257,共8页Computer Engineering

基　　金：上海市科委重点项目(14DZ1206302)

摘　　要：在低信噪比及非平稳的噪声环境下,传统基于特征的语音端点检测方法检测正确率低、稳定性差。为此,提出一种新的语音端点检测方法。通过对含噪语音进行谱减法降噪,提取谱减后的语音信号与前导无话帧的M FCC倒谱距离特征,计算均匀子带频带方差特征,并对阈值进行动态更新,利用双参数双门限法对带噪语音进行端点判定。实验结果表明,与基于DWT-MFCC倒谱距离、基于谱减法和均匀子带频带方差的端点检测方法相比,该方法具有较高的检测正确率及较低的漏检率与误检率。In view of low signal-to-noise ratio and non-stationary noise environment,the traditional methods based on feature detection have the low accuracies and poor stabilities.To solve this problem,this paper proposes a new speech endpoint detection method.The spectral subtraction method is used to reduce noise.Then the MFCC cepstrum distance features of the speech signal after spectral subtraction and the leading silent frame are extracted,and also the frequency variance characteristics of the uniform sub-band are extracted.And the dynamic threshold updating mechanism is used to detect the noisy speech with two-parameter double threshold method.Experimental results show that,compared with the method based on DWT-MFCC cepstrum distance and the method based on spectral subtraction and uniform sub-band variance,the proposed method has a higher accuracy and the lower miss rate and error rate.

关 键 词：端点检测 谱减 MFCC倒谱距离 均匀子带方差 动态阈值更新 

分 类 号：TP37[自动化与计算机技术—计算机系统结构]