首页 | 本学科首页   官方微博 | 高级检索  
     

一种基于模糊规则的鲁棒语音识别方法
引用本文:张军,章熙春,曹燕,韦岗.一种基于模糊规则的鲁棒语音识别方法[J].电路与系统学报,2006,11(5):96-100.
作者姓名:张军  章熙春  曹燕  韦岗
作者单位:华南理工大学,电信学院,广东,广州,510640
基金项目:国家自然科学基金,华南理工大学校科研和教改项目
摘    要:本文在丢失数据技术与声学后退技术的基础上,提出了一种基于模糊规则的鲁棒语音识别方法,首先根据先验知识或假定建立特征分量的可靠程度与其概率分布之间的模糊规则,识别时观察矢量的输出概率由一个基于规则的模糊逻辑系统来得到,并针对倒谱识别系统给出了一种具体的实现方法.实验结果表明,所提识别方法的性能显著优于丢失数据技术和声学后退技术.

关 键 词:语音识别  鲁棒性  模糊  丢失数据技术  声学后退技术
文章编号:1007-0249(2006)05-0096-05
收稿时间:2004-06-21
修稿时间:2004-12-03

A robust speech recognition approach based on fuzzy-rule
ZHANG Jun,ZHANG Xi-chun,CAO Yan,WEI Gang.A robust speech recognition approach based on fuzzy-rule[J].Journal of Circuits and Systems,2006,11(5):96-100.
Authors:ZHANG Jun  ZHANG Xi-chun  CAO Yan  WEI Gang
Abstract:This paper proposes a new fuzzy-rule-based approach for noisy speech recognition based on the missing data and acoustic backing-off techniques. First for each feature component the fuzzy rules between its reliability level and distribution are built using prior knowledge or assumptions. Then in the recognition process the output probability of an observation is calculated by a rule based fuzzy logic system. The proposed approach is implemented on a cepstral speech recognizer. Experimental results show that it significantly outperforms the missing data and acoustic backing-off techniques.
Keywords:speech recognition  robustness  fuzzy  missing data technique  acoustic backing-off technique
本文献已被 CNKI 维普 万方数据 等数据库收录!
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号