基于小波分析的大词汇汉语连续语音识别系统鲁棒性的研究  

A Study on Robustness of Large Vocabulary Chinese Continuous Speech Recognition System Based on Wavelet Analysis

作　　者：颜龙[1] 刘刚[1] 郭军[1] 

机构地区：[1]北京邮电大学信息工程学院,北京100876

出　　处：《中文信息学报》2006年第2期60-65,共6页Journal of Chinese Information Processing

基　　金：教育部跨世纪人才基金资助项目;教育部重点项目(02029)

摘　　要：本文提出一种基于小波分析的大词汇汉语连续语音识别的方法,即采用一维小波变换将原始语音信号进行五层小波分解,然后对各层小波系数进行重构,得到五层语音信号,分别对各层语音信号进行训练,得到各层的声学模型,然后结合语言模型对各层声学模型的性能进行测试。通过对纯净语音和带噪语音的各层重构语音数据进行测试。结果表明对于含有高斯白噪声的带噪语音,该方法能使系统性能有所提高,但对于粉红噪声,该方法效果不明显。对于含有真实环境噪声的带噪语音,该方法能获得比基线系统更好的性能。In this paper wavelet decomposition is used to decompose speech signal into five levels. The wavelet coefficients of each part were reconstructed. Because different frequencies of the speech signal have different influence on the performance of the system, the acoustic model of each level was trained and tested. The experimental results show that the method of this paper is effective on gauss white noise and real environmental noise. However it is not effective on pink noise.

关 键 词：计算机应用 中文信息处理 大词汇连续语音识别 小波分析 声学模型 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]