摘 要：在封闭的环境内，声源发出的语音信号不可避免会因为房间冲激响应产生一定的混响，造成语音质量和辨识度下降，给通信的对方带来一定的困难。因此，高效的语音去混响算法变得特别引人注目，研究过程中最常用的方法就是多通道线性预测自适应去混响算法，文章将详细介绍该算法，提出新的时变遗忘因子，控制方案并改进了QR-RLS算法，不仅提高了去混响的能力，还提高了算法的稳定性。
关键词：去混响；多通道线性预测；自适应滤波；最小二乘法；QR 分解；时变遗忘因子
DOI:10.19850/j.cnki.2096-4706.2022.08.027
中图分类号：TP18 文献标识码：A 文章编号：2096-4706（2022）08-0095-03
Research on QR-RLS De-Reverberation Model with Variable Forgetting Factor
FENG Tao, SUN Yiran
(Nanjing University of Information Science and Technology, Nanjing 210044, China)
Abstract: In a closed environment, the speech signal emitted by the sound source will inevitably produce a certain reverberation due to the room impulse response, which will cause the speech quality and recognition to decline, and bring certain difficulties to the other side of communication. Therefore, efficient speech de-reverberation algorithms have become particularly attractive. The most commonly used method in the research process is the multi-channel linear prediction adaptive de-reverberation algorithm. This paper will introduce the algorithm in detail and propose a new time-varying forgetting factor, control scheme and improve the QR-RLS algorithm. It not only improves the ability of de-reverberation, but also improves the stability of the algorithm.
Keywords: de-reverberation; multi-channel linear prediction; adaptive filtering; least square procedure; QR decomposition; timevarying forgetting facto
参考文献：
[1] JUKIĆ A，WATERSCHOOT T，GERKMANN T，et al. Multi-Channel Linear Prediction-Based Speech Dereverberation With Sparse Priors [J].IEEE/ACM Transactions on Audio，Speech，and Language Processing，2015，23（9）：1509-1520.
[2] 云霞 . 基于 Mel 倒谱和 Bark 谱失真距离的汉语音质客观评价研究 [D]. 成都：西南交通大学，2004.
[3] WU M Y，WANG D L. A two-stage algorithm for onemicrophone reverberant speech enhancement [J].IEEE Transactions on Audio，Speech，and Language Processing，2006，14（3）：774-784.
[4] 李毅 . 自适应滤波及滤波算法研究 [D]. 西安：西北工业大学，2003.
[5] XIANG T. Study on Speech Enhancement with Reverberation [D].Nanjing：Nanjing University，2020.
作者简介：冯涛（2000.11—），男，汉族，江苏无锡人，本科在读，研究方向：语音去混响；孙怡然（2001.01—），男，汉族，江苏淮安人，本科在读，研究方向：语音去混响。