基于MFCC相似度和谱熵的端点检测算法  

Endpoint detection algorithm based on MFCC similarity and spectrum entropy

作　　者：邓瑞[1] 肖纯智[1] 高勇[1] 

机构地区：[1]四川大学电子信息学院,四川成都610065

出　　处：《现代电子技术》2013年第21期67-69,共3页Modern Electronics Technique

摘　　要：为提高低信噪比环境下语音端点检测的准确率,提出了一种基于Mel倒谱参数相似度和谱熵的端点检测算法。首先,提取语音帧的的Mel频率倒谱参数,将前十帧声信号作为背景噪声,然后计算每一帧语音和噪声MFCC的相关系数距离,结合MFCC相似距离与谱熵做综合判决。实验结果表明,在低信噪比环境下此方法相对谱熵法能够提高检测准确率。In order to improve the accuracy of Voice Activity Detection （VAD） in the low signal-to-noise ratio environ-ment,an algorithm based on Mel frequency cepstrum coefficient（MFCC）similarity and spectrum entropy is proposed. The MF-CC is extracted from each speech frame,and the first ten frames of sound signal are taken as the background noises,then the MFCC correlation coefficient distance of each frame of the voice and noise is calculated,the comprehensive judgment is made in combination with the MFCC similarity distance and spectrum entropy. The experiment result shows that the new algorithm can improve the accuracy of the detection under low SNR environments compared with spectrum entropy method.

关 键 词：语音信号处理 端点检测 Mel频率倒谱参数 相关系数 谱熵 

分 类 号：TN912.334[电子电信—通信与信息系统]