短时傅里叶变换域最优非因果滤波器和滤波矩阵降噪算法  

Noise reduction using optimal non-causal filters and rectangular filtering matrices in short-term Fourier transform domain

作　　者：王向辉 李梅 田旭华 王姣 谭歆[1] 路东东 WANG Xiang-hui;LI Mei;TIAN Xu-hua;WANG Jiao;TAN Xin;LU Dong-dong(School of Electronic Information and Artificial Intelligence,Shaanxi Joint Laboratory of Artificial Intelligence,Shaanxi University of Science&Technology,Xi′an 710021,China;Beijing Institute of Spacecraft Environmental Engineering,Beijing 100094,China)

机构地区：[1]陕西科技大学电子信息与人工智能学院陕西省人工智能联合实验室,陕西西安710021 [2]北京卫星环境工程研究所,北京100094

出　　处：《陕西科技大学学报》2024年第2期164-173,197,共11页Journal of Shaanxi University of Science & Technology

基　　金：国家自然科学基金项目(62101313);陕西省科技厅自然科学基础研究计划面上项目(2023-JC-YB-484);陕西省重点研发计划项目(2020GY-091);陕西省西安市高校院所科技人员服务企业项目(22GXFW0004);陕西科技大学博士科研启动基金项目(2020BJ-049)。

摘　　要：在现有短时傅里叶变换域单通道降噪算法考虑帧间相关性的基础上,为提高算法的灵活度以及滤波器的性能,提出了一系列最优非因果滤波器和滤波矩阵.具体地,通过引入非因果概念,设计出一组单通道非因果降噪滤波器,以实现同时利用过去帧和未来帧信息进行噪声抑制的目的.并在此基础上,推导出了一组最优单通道滤波矩阵,以降低算法的复杂度.将非因果滤波器以及滤波矩阵的思想相结合,进一步提出了一组单通道非因果滤波矩阵.随后,将所提算法扩展至多通道场景中,设计出了一组最优多通道非因果降噪滤波矩阵,以达到更好的降噪效果.最后,对所提算法进行了仿真实验.实验结果表明,与传统算法相比,本文提出的算法可有效改善带噪信号(输入信号)的信噪比,提升语音质量.In order to increase the flexibility of the algorithm as well as improve the performance of the filter,a set of optimal non-causal filters and filtering matrices are developed based on the existing single-channel noise reduction algorithms in Short-Time Fourier Transform(STFT)domain which takes the interframe correlation into account.Specifically,this paper designs a set of single-channel non-causal filters by introducing the non-causal concept to achieve the purpose of noise reduction by exploiting both the past and future information.Based on that,several optimal single-channel filtering matrices are also derived to make the computational burden of the algorithms lower.Combining the aforementioned ideas,a set of single-channel non-causal filtering matrices are deduced.Furthermore,the proposed algorithms are generalized to the multi-channel case to achieve a better performance.Finally,the algorithms developed previously are investigated through simulations.The results show that the proposed algorithms can effectively improve the signal-to-noise ratio of the noisy signal(input signal)and enhance the speech quality compared with the traditional algorithms.

关 键 词：语音降噪 非因果 滤波矩阵 短时傅里叶变换域 单通道 多通道 

分 类 号：TN912[电子电信—通信与信息系统]