基于小波包与分形的语音特征提取  

The Method of Speech Feature Extraction Based on Wavelet Packet and Fractal Dimensiom

作　　者：陈春辉[1] 

机构地区：[1]华南师范大学计算机学院,广州510631

出　　处：《软件导刊》2012年第6期159-161,共3页Software Guide

摘　　要：针对语音信号的非平稳特性,传统的短时分析技术容易丢失信息的现状,提出了一种利用小波包变换和分形技术对语音信号进行特征提取。用小波包变换对每一帧语音信号进行多分辨分析,这样可以得到每个频带的小波分解值,然后计算出每个频带的分形维特征,最后使用动态时间规整进行识别。实验证明,在低信噪比的情况下,仍有较高的说话人识别的效果。According to the Non-stationary characteristics of the speech signal,it is easy to lose the present situation of the information to the traditional short analysis technique,and puts forward a kind of wavelet packet transform and fractal dimension technology to speech signal feature extraction.Using wavelet packet transform to each frame speech signal multi-resolution analysis,this can get each band of wavelet decomposition value,and then figure out each band of the fractal dimension characteristic,finally use dynamic time warping to recognize.Experimental proof： in low SNR circumstance,still have higher speaker recognition effect.

关 键 词：小波包 分形 语音特征 动态时间归整 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]