摘要
利用话者识别原理和语音数字信号处理技术对人声建模方法进行研究,建立了基于GMM模型的VDR环境下的人声识别基准系统;从分析影响人声识别率因素的角度出发,指出传统算法的不足,并提出一种基于近似熵的语音端点检测算法。理论分析和实验结果证明:新算法能有效屏蔽大动态冲击性噪声,解决了语音的虚检现象,并且在低信噪比0 dB情况下的识别率提升66%。
Human voice modeling method is researched,using speaker recognition principles and digital signal processing technology,basic system of voice recognition based on GMM model in VDR is established.From the perspective analysis of affecting on voice recognition ratio,the lack of traditional algorithms is pointed out and an endpoint detection algorithm of speech based on approximate entropy is made.Theoretical analysis and experimental results confirmed that the new algorithm can effectively shield large dynamic impact noise,problem of voice virtual check is solved,and increased voice recognition rate of 66 % in low SNR 0 dB.
出处
《传感器与微系统》
CSCD
北大核心
2011年第6期61-64,70,共5页
Transducer and Microsystem Technologies
基金
黑龙江省教育厅2009年度科学技术研究项目(11541304)
关键词
人声识别
话者识别
航行数据记录仪
近似熵
端点检测
voice recognition
speaker recognition
voyage data recorder
approximate entropy
endpoint detection
