单通道语音增强技术的研究现状与发展趋势  

Research status and development trend of single-channel speech enhancement methods

作　　者：许春冬[1] 徐琅 周滨 凌贤鹏 XU Chundong;XU Lang;ZHOU Bin;LING Xianpeng(School of Information Engineering,Jiangxi University of Science and Technology,Ganzhou 341000,Jiangxi,China)

机构地区：[1]江西理工大学信息工程学院,江西赣州341000

出　　处：《江西理工大学学报》2020年第5期55-64,共10页Journal of Jiangxi University of Science and Technology

基　　金：国家自然科学基金资助项目(11864016);江西省文化艺术科学规划项目一般项目(YG2017384)。

摘　　要：语音增强技术在语音信号处理领域得到了充分重视和广泛研究,其目的是降低噪声对语音信号的影响,提升目标语音信号的质量。文章首先分析了语音增强基本模型、噪声类型以及语音质量评价方法,其次详细介绍了几种传统的语音增强方法以及监督性单通道语音增强方法。重点介绍了几种代表性的基于深度神经网络的语音增强方法,包括基于DNN、CNN、RNN、LSTM、GAN等网络的语音增强方法。最后总结了几种常用语音增强方法的优缺点,并根据深度学习和语音处理的发展现状,分析了语音增强技术所面临的挑战和发展趋势。Speech enhancement technology has received full attention and extensive research in the field of speech signal processing.Its purpose is to reduce the influence of noise on speech signals and improve the quality of target speech signals.Firstly,the article analyzes the basic model of speech enhancement,noise types and speech quality evaluation methods,and then introduces several traditional speech enhancement methods and supervised single-channel speech enhancement methods in detail.It focuses on several representative speech enhancement methods based on deep neural networks,including speech enhancement methods based on DNN,CNN,RNN,LSTM,GAN and other networks.Finally,the advantages and disadvantages of several commonly used speech enhancement methods are summarized,and the challenges and development trends of speech enhancement technology are analyzed according to the development status of deep learning and speech processing.

关 键 词：单通道 语音增强 深度神经网络 机器学习 

分 类 号：TN912.35[电子电信—通信与信息系统]