李　俊　周　萍　景新幸

1(桂林电子科技大学电子工程与自动化学院　广西 桂林 541004)

2(桂林电子科技大学信息与通信学院　广西 桂林 541004)



车载环境下语音增强的研究

李俊1周萍1景新幸2

1(桂林电子科技大学电子工程与自动化学院广西 桂林 541004)

2(桂林电子科技大学信息与通信学院广西 桂林 541004)

摘要针对车载环境下语音系统受到外界强噪声的干扰而导致识别精度降低以及通信质量受损的问题，提出一种自适应MMSE-LSA估计与TEO(Teager Energy Operator)能量端点检测相结合的语音增强算法。TEO端点检测可以将语音分为语音段和非语音段，从而在噪声估计时可以更好地跟踪噪声的变化，得到更加准确的先后验信噪比，使增强后的语音最大限度地接近纯净语音，而且对车载噪声的增强效果比其他噪声更好。在车载环境中进行实验，结果显示该方法与MMSE-LSA以及传统的谱减法相比，提高了输出信噪比，减弱了音乐噪声，在可懂度和清晰度方面均具有优势。

关键词车载噪声信噪比TEO端点检测语音增强

RESEARCH ON SPEECH ENHANCEMENT ALGORITHM UNDER ON-BOARD ENVIRONMENT

Li Jun1Zhou Ping1Jing Xinxing2

1(School of Electronic Engineering and Automation,Guilin University of Technology,Guilin 541004,Guangxi,China)2(School of Information and Communication,Guilin University of Technology,Guilin 541004,Guangxi,China)

AbstractVoice system could be disturbed by external strong noise under on-board environment, which will lead to decline in accuracy of recognition and damage of communication quality. In order to solve this problem, we proposed a speech enhancement algorithm combining adaptive MMSE-LSA estimation and TEO (Teager energy operator) endpoint detection. According to TEO endpoint detection method, speech can be divided into voice segment and non-voice segment so that in noise estimation the changes of noise can be better tracked and more accurate priori SNR and posterior SNR can be gained as well, this makes the enhanced speech sufficiently approach the original speech. In addition, TEO endpoint detection has a much better effect in enhancing vehicle noise than any other noises. Experiment was carried out under on-board environment, results showed that this method, compared with MMSE-LSA estimation and traditional spectral subtraction, improved the output SNR, reduced the music noise, and had the advantages in both intelligibility and clarity.

KeywordsOn-board noiseSignal-to-noise ratioTEOEndpoint detectionSpeech enhancement

0引言

在汽车内部由于动力系统的运行产生的齿轮啮合，高速行驶时车轮与地面和空气的摩擦以及其他振动源共同形成了车载噪声[3]。车载语音通信难免会受到噪声的干扰，带有很强的背景噪声的语音信号很难被车载语音系统识别，必须对车载语音进行增强处理，消除背景噪声，提高汽车中语音通信的质量和车载系统的识别率。

传统的谱减法、维纳滤波法、最小均方差估计法在去除背景噪声方面有很好的效果，但是有的时候会造成语音的失真或者产生强烈的音乐噪声[2,4]。……