低信噪比环境下的麦克风阵列语音识别算法研究  

Speech Recognition Based on Microphone Array in Low SNR

作　　者：王群[1] 曾庆宁[1] 郑展恒[1] 

机构地区：[1]桂林电子科技大学信息与通信学院,桂林541004

出　　处：《科学技术与工程》2017年第31期101-107,共7页Science Technology and Engineering

基　　金：国家自然科学基金(61461011);教育部重点实验室2016年主任基金(CRKL160107)资助

摘　　要：针对在低信噪比环境下语音增强对语音识别率的提升不明显的问题,提出一种用在识别系统前端的麦克风阵列增强算法。该阵列增强算法基于相干滤波和频率带宽波束形成后置改进维纳滤波器。首先将采集到的阵列信号,求相邻通道间的相关函数,利用阵元间信号的相关性进行初始噪声抑制,然后利用频域宽带最小方差无畸变响应(MVDR)通过对目标声源信息的获取,保留目标声源方向的信号并抑制其他方向的信号干扰,再通过改进的维纳滤波器去除噪声残留提升语音可懂度,最后用梅尔频率倒谱系数(MFCC)和隐马尔科夫模型(HMM)对增强后的语音信号做特征提取并识别。仿真过程模仿双耳采集数据,结果表明该语音增强方法在低信噪比环境下获得较好的增强效果,能有效的提高低信噪比环境下的语音识别率。For improving the accuracy rate of speech recognition in low SNR environment,a microphone array speech enhancement based on coherent filter and MVDR-wiener is presented. Improving the gathered speech signals' SNR by the coherent filter is used relative signals information between microphones firstly. Then obtaining the information of target speech by the broadband minimum variance distortionless response algorithm( MVDR) to keep the desired direction of the signal and the other direction of the interference signal will be suppressedand. The improved wiener filter which can get better intelligibility was utilized to process the enhanced signal. At the last,an HMM based system with MFCC as input feature is used to recognitize the enhanced speech. Binaural recording data were used during the simulation. Results show that the proposed method can effectively remove the background noise and reduce the residual noise,significantly increase the accuracy of the automatic speech recognition in noisy environment.

关 键 词：语音增强 相干滤波 最小方差无畸变响应 改进维纳滤波 语音识别 双耳听觉 

分 类 号：TN912.34[电子电信—通信与信息系统]