基于多任务稀疏表达的二元麦克风小阵列话音增强算法

发布时间：2018-06-17 20:30

本文选题：麦克风小阵列 + 话音增强　；参考：《通信学报》2014年02期

【摘要】：针对常规二元麦克风小阵列话音增强算法通常需要话音活动检测技术支持,并且难以有效抑制第一帧含目标信号的噪声。提出了一种基于多任务稀疏表达的二元麦克风小阵列话音增强算法,首先利用字典学习方法分别获得目标信号和噪声信号的过完备字典,然后利用2 1???混合范数对信号在其字典上的表示系数进行正则化稀疏约束,使得2个阵元接收到信号中的噪声信号被抑制,而话音信号尽量保持不变,从而达到话音增强的目标。仿真和实验数据表明,无论开始位置是否含有目标话音信号,所提出的非话音活动检测支持的二元麦克风小阵列话音增强算法均能有效实现话音增强的目标。
[Abstract]:For the conventional dual microphone small array voice enhancement algorithms usually need voice activity detection technology support and it is difficult to effectively suppress the noise of the first frame with the target signal. In this paper, a multi-task sparse representation based speech enhancement algorithm for binary microphone array is proposed. Firstly, the over-complete dictionaries of target signal and noise signal are obtained by using dictionary learning method, and then the over-complete dictionary of target signal and noise signal is obtained by using 21? The mixed norm regularizes the representation coefficients of the signals in their dictionaries so that the noise signals in the signals received by two array elements are suppressed while the voice signals are kept unchanged so as to achieve the goal of speech enhancement. Simulation and experimental data show that the proposed speech enhancement algorithm supported by non-speech activity detection can effectively achieve the target of speech enhancement regardless of whether the starting position contains the target voice signal or not.
【作者单位】：浙江大学计算机科学与技术学院;浙江万里学院智能控制技术研究所;
【基金】：国家自然科学基金资助项目(61171151) 国家重点基础研究发展计划(“973”计划)基金资助项目(2012CB316400) 国家科技支撑计划基金资助项目(2011BAD24B03)~~
【分类号】：TN912.35

【相似文献】