噪声环境下统计分类语音检测的自适应算法  

Adaptive Algorithm of Speech Detection Based on Statistical Classifica tion in the Presence of Noise

作　　者：田野[1] 王作英[1] 陆大金[1] 

机构地区：[1]清华大学电子工程系,北京100084

出　　处：《计算机工程与应用》2002年第1期14-15,38,共3页Computer Engineering and Applications

基　　金：国家863高技术研究发展计划项目(编号:863-306-ZD03-02-1);清华大学985重大项目"人机自然语言交互技术"(编号:985校-22-攻关-06)

摘　　要：可靠的语音检测算法是稳健语音识别系统所必需的。该文针对统计分类语音算法对训练数据的依赖问题,提出自适应算法在线动态更新分类模型。实验结果表明在训练和识别环境失配情况下,自适应算法可以明显提高语音检测的稳健性。在稳态的噪声环境下,语音检测的正确率有3%-10%的提高;在时变的噪声环境下,语音检测的正确率有7%-28%的提高。The reliability of speech detection should be taken into consideration for speech recognition in noise.Adap-tive algorithm of speech detection based on statistical classification is presented to reduce the dependence of the train-ing data.The probability distribution of speech and noise model is updated online.Experiments show that the adaptive al-gorithm is efficient.The correct detection rate is increased about 3%-10%under single noise environment and about 7%-28%under time-various noise environments.

关 键 词：语音检测 统计分类 噪声环境 自适应算法 

分 类 号：TP391.42[自动化与计算机技术—计算机应用技术]