嵌入式语音识别在混响环境中的信号增强方法  

Signal enhancement method of embedded speech recognition in reverberation environment

作　　者：郭威[1] 申利民[1] 谭云福[1] 陶凤莉[1] 

机构地区：[1]燕山大学信息科学与工程学院,河北秦皇岛066004

出　　处：《计算机应用研究》2010年第12期4565-4567,4581,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(60873008)

摘　　要：针对在混响环境中使用嵌入式平台的语音识别系统,提出了一种高效实时的在混响环境下带干扰噪声的语音信号增强方案。该方法以阵列麦克风为前端语音拾取设备,对每个麦克风之间采样得到的语音信号进行多径分集接收处理,通过分析语音信号之间的相位关系,多波束形成,对相干信号延时处理并加权合并提高信噪比以实现对采集得到的语音信号的增强处理;通过调整权值矩阵滤除非语音频段信号和噪声,进而进一步降低可能引入的噪声污染。此外,该方案的算法强度较小,便于在嵌入式平台上使用。In view of useing the embedded platform＇s speech recognition system in the reverberation environment,this paper proposed one kind of highly effective real-time program under reverberation environment the belt interference noise voice signal enhancement. This method took the array microphone as the front end speech collection equipment,carried on multi-diameter diversity reception to the voice signal which the sampling obtains to each between microphone. Analyzed the phase relation between the voice signal,formed the multi-beam,delayed,weighted and combined coherent signals,and improved the signal-to noise ratio to enhance the voice signal obtained by gathering. Adjusted weighted matrix filters,only if the speech frequency band signal and the noise,then further reduced the noise pollution which possibly introduced. In addition,the proposed method has small algorithm intensity,and is convenient to use in the embedded platform.

关 键 词：阵列麦克风 语音增强 相位差 波束形成 多径传播 

分 类 号：TP912.35[自动化与计算机技术]