一种鲁棒性音素分段算法

A Kind of Robust Speech Segment Algorithm

基于小波变换的鲁棒性因素分段算法的基本思想是在运用传统的参数滤波方法进行音素分段之前首先将语音信号在小波域中进行滤波,提出对听觉感知有效的语音分量,然后用传统的参数滤波方法进行分段。参数滤波是以一个变化的参数对信号进行滤波,得到信号在不同频带中的分量,可以证明,若滤波参数以一定的规律变化,则这些滤波分量的一阶自相关表示了信号的相关结构。利用新方法进行分段并测试其鲁棒性,实验证明新方法分段效果好且鲁棒性强,是一种有效的音素分段算法。

The main idea of the algorithm is to filter the speech signal in wavelet domain before using the traditional parametric filtering method, which can abstract the effective speech signals that are important to human ear. The parametric filtering method is motivated by the fact that the correlation structure of a stationary signal can be characterized by the signature of certain output statistics from a judiciously designed filter bank. This new method is used to detect the speech segmentations and to analyze the robustness. It is proved that this method has good ability and high robustness, and that it is an effect method in speech segmentation.