汽车噪声中自动语音的识别技术  

Techniques for Automatic Speech Recognition in Car Noise

作　　者：韦晓东[1,2] 朱杰[1,2] 胡光锐[1,2] 

机构地区：[1]上海交通大学与贝尔实验室通信与网络联合实验室 [2]上海交通大学电子工程系

出　　处：《上海交通大学学报》1998年第10期10-13,共4页Journal of Shanghai Jiaotong University

摘　　要：汽车中的话音拨号系统是自动语音识别技术的应用热点．自动语音识别系统是一个基于训练的系统．在汽车噪声中，由于实际应用环境与形成系统参数的训练环境的失配，传统语音识别系统的性能会大幅度地下降，从而无法实用．为了提高语音识别系统在特定环境下的识别率及实用性，首先根据汽车环境中语音的失真模型分析了系统性能下降的原因，然后针对加性汽车噪声与信道失真对系统的影响，讨论了在汽车噪声中改善语音识别系统性能的方法．提出了在识别系统中用基于子带的语音增强算法和倒谱均值相减算法相结合的方法．对大量的多人连续数字串语音的识别实验表明，这一方法大大提高了系统在汽车噪声环境中的识别率，它还可以简便、实时的实现，具有一定的实用性．Voice dialing system is a hot application issue for automatic speech recognition(ASR). Speech recognition system is always based on training, so its performance degrades significantly when there is a mismatch between testing and training. To improve the robustness of ASR in car noise, this paper analyses the main reason why the performance degrades. For the effect of additional noise and channel distortion, some techniques to improve ASR in car noise are described. The main two methods——speech enhancement based on subband filtering and cepstral mean subtraction are proposed. Both of them can be real time implemented to fulfill the requirement of speech recognition system in cars. The experiments show that higher accuracy recognition rate can be achieved in car noise.

关 键 词：语音识别 语音增强 汽车噪声 

分 类 号：U491.91[交通运输工程—交通运输规划与管理] TN912.34[交通运输工程—道路与铁道工程]