基于分形维数的带噪语音端点检测  

Endpoint detection of noisy speech based on fractal dimension

作　　者：贾亮 尹伊 杨慧超 

机构地区：[1]沈阳航空航天大学电子信息工程学院,沈阳110136

出　　处：《沈阳航空航天大学学报》2017年第5期63-67,共5页Journal of Shenyang Aerospace University

摘　　要：噪声环境下语音端点检测的难度会大大提高,应用分形维数法对带噪语音进行了分析,利用分形的性质,可以更好地分析语音特性,根据分形维数的不同可以实现语音段与噪音段的分割,实现带噪语音的起止点检测。通过对比分形维数的端点检测法与传统的谱熵端点检测法,针对在信噪比为0d B与5d B的语音信号进行仿真,结果表明,分形维数的方法能更好地分割语音段与噪声段。The difficulty of speech endpoint detection in noisy environment w ill be greatly improved,The application of fractal method to analyze the noisy speech using fractal properties,can better analyze the speech characteristics,according to the different fractal dimension can realize the speech and noise segment,realize noisy speech starting and ending point detection. The spectral entropy endpoint detection method of endpoint detection method compared w ith the traditional fractal dimension,according to the simulation,0 d B and 5 d B than the speech signal noise. The results show that the method of fractal dimension can better segmentation of speech and noise.

关 键 词：分形维数 语音端点检测 谱熵 带噪语音 

分 类 号：TN912.3[电子电信—通信与信息系统]