一种采用旁瓣增强的麦克风阵列抗混响算法

A Microphone-array De-reverberation Method Using Sidelobe Enhancement

在音频/视频会议、人机交互、语音识别等领域,严重的混响干扰导致麦克风阵列语音处理性能急剧下降.针对现有逆滤波等抗混响方法需要获得准确的房间传输响应,而波束形成方法抗混响性能有限的问题,基于广义旁瓣抵消器(generalized sidelobe canceller,GSC)结构提出一种采用旁瓣增强的麦克风阵列抗混响算法(Sidelobeenhanceing reverberation mitigation algorithm for microphone array,SERM).该算法首先进行波束形成,初步获得增强的直达语音信号,并对旁瓣获取的混响分量进行自适应迭代增强,再将旁瓣迭代增强的误差信号作为参考噪声进行自适应噪声抵消,最终输出抗混响语音.实验结果表明,在混响环境下该方法能有效改善麦克风阵列的语音信号质量.

In many practical applications such as audio/video conferences,man-machine interfaces,speech recognitions,the reverberation leads to significant degradation of microphone array speech processing.However,while the conventional de-reverberation methods,such as inverse-filtering method,need to obtain accurate room transmission responses,the classic beam-forming method only yields limited performance.Based on the Generalized Sidelobe Canceller（GSC）structure,this paper proposes a novel microphone-array de-reverberation method to achieve sidelobe exploitation.First,we use the contemporary beam-forming method to initially obtain the direct speech signals.Then we adopt the adaptive iterative algorithm to explore reverberation components in sidelobes for further enhancements.Finally,the error of the sidelobe enhancement is adopted as the reference noise of adaptive noise cancellation to achieve the de-reverberation.Experimental results show that the proposed method effectively improves the quality of speech signal under the reverberation environment.