一种改进的语音基频轮廓提取算法  

A New Algorithm to Extract the Fundamental Frequency of Speech

作　　者：江太辉[1] 

机构地区：[1]五邑大学电子与信息工程系,广东江门529020

出　　处：《五邑大学学报（自然科学版）》2002年第2期27-30,34,共5页Journal of Wuyi University(Natural Science Edition)

基　　金：广东省自然科学基金资助项目(000872)

摘　　要：在说话人识别课题中,需要提取精确的基频轮廓作为特征参量,但通常使用传统的固定窗口长度的短时自相关函数算法并不能够适应各种不同人的发音. 论文使用了一种可变窗口长度的改进算法,实验表明它有更强的适应能力. 文中还介绍了一种改进的基频“野点”平滑算法,它能够在去除野点的同时,更好地保存正确点的信息.The fundamental frequency of speech is an important parameter in speaker recognition. The traditional method uses constant length frame window. It cannot adapt to different people with different tones. A new algorithm that uses alterable length frame window is proposed in this paper. Experiments show that it has better performance than the old ones. This paper also introduces an improved fundamental frequency-contour smoothing filter. It can preserve more valid information than the traditional algorithms.

关 键 词：语音基频轮廓提取算法 语音信号处理 平滑滤波 语音识别 说话人识别 提取原理 

分 类 号：TN912.34[电子电信—通信与信息系统]