低信噪比环境下的麦克风阵列语音识别算法研究

发布时间：2018-05-29 15:22

本文选题：语音增强 + 相干滤波　；参考：《科学技术与工程》2017年31期

【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。
[Abstract]:A microphone array enhancement algorithm used in the front end of the recognition system is proposed to solve the problem that the speech recognition rate is not obviously enhanced by speech enhancement in low SNR environment. The array enhancement algorithm is based on coherent filtering and frequency bandwidth beamforming post-improved Wiener filter. Firstly, the correlation function between adjacent channels is obtained from the collected array signals, and the initial noise suppression is carried out by using the correlation between array elements. Then, the target source information is obtained by using the wideband minimum variance distortionless response (MVDRR) in frequency domain. The signal in the direction of the target sound source is retained and the signal interference in other directions is suppressed, and then the speech intelligibility is improved by removing the residual noise from the improved Wiener filter. Finally, Mel Frequency Cepstrum coefficient (MFCC) and Hidden Markov Model (HMMM) are used to extract and recognize the features of the enhanced speech signal. The simulation results show that the proposed method achieves better enhancement effect in low SNR environment and can effectively improve the speech recognition rate in low SNR environment.
【作者单位】：桂林电子科技大学信息与通信学院;
【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
【分类号】：TN912.34

【相似文献】