变分自编码器在环境噪声消除中的应用研究  

Research on the Application of Variational Autoencoder in Environmental Noise Cancellation

作　　者：李红玲 LI Hongling(Guangxi Electronic Senior Technical School,Nanning 530031,China)

机构地区：[1]广西电子高级技工学校,广西南宁530031

出　　处：《电声技术》2024年第8期105-107,共3页Audio Engineering

摘　　要：针对语音信号中的环境噪声影响问题,研究一种基于变分自编码器(Variational Auto-Encoder,VAE)的噪声消除方法,并进行了相关实验验证。首先,深入研究VAE的基本原理;其次,引入L_2正则化方法优化VAE,以提高模型的泛化能力和健壮性;最后,利用ESC-50和VCTK数据集构建包含环境噪声的语音信号数据集,并在此基础上进行了一系列实验。实验结果表明,所提方法能够有效降低环境噪声对语音信号的影响。This article focuses on the impact of environmental noise in speech signals,and studies a noise cancellation method based on Variational Auto-Encoder(VAE),and conducts relevant experimental verification.Firstly,this article delves into the basic principles of VAE;Subsequently,the L,regularization method was introduced to optimize the VAE,in order to improve the model's generalization ability and robustness;Subsequently,a speech signal dataset containing environmental noise was constructed using the ESC-50 and VCTK datasets,and a series of experiments were conducted on this basis.The experimental results show that the proposed method can effectively reduce the impact of environmental noise on speech signals.

关 键 词：变分自编码器(VAE) 正则化 环境噪声 去噪 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TN641[自动化与计算机技术—控制科学与工程]