基于生成对抗网络的数字音频信号多声道增强方法  

Digital audio signal multichannel enhancement methodbased on generative adversarial network

作　　者：胡嘉欣 田军[1] HU Jiaxin;TIAN Jun(North University of China,Taiyuan 030051,China)

机构地区：[1]中北大学,山西太原030051

出　　处：《现代电子技术》2023年第19期41-44,共4页Modern Electronics Technique

摘　　要：基于生成对抗网络的数字音频信号多声道增强方法,提升数字音频清晰度、降低噪声对其的干扰。以结合傅里叶的双边语谱图滤波算法初步实现数字音频多声道信号去噪;构建基于对抗神经网络的数字音频信号多声道增强模型,将去噪后的数字音频多声道信号作为该模型的输入数据,输出增强的多声道数字音频信号,完成数字音频信号多声道增强。实验结果表明:该方法增强后的数字音频信号传输效果较好,同时信噪比和音频客观评判基准较高,且去噪后语谱图中噪点以及模糊区较少,能够有效去除信号中的噪声,提升数字音频信号质量。A digital audio signal multi⁃channel enhancement method based on generative adversarial network(GAN)is studied to enhance digital audio definition and reduce noise interference.The denoising of digital audio multi⁃channel signals is realized preliminarily by bilateral spectrogram filtering algorithm combined with Fourier transform.A digital audio signal multi⁃channel enhancement model based on adversarial neural network is constructed.The denoised digital audio multi⁃channel signals are taken as the input data for the model to output the enhanced multi⁃channel digital audio signals,so as to complete multi⁃channel enhancement of digital audio signals.The experimental results show that the transmission effect of the digital audio signals enhanced by this method is good;its signal⁃to⁃noise ratio(SNR)and audio objective evaluation criteria are relatively high;after denoising,there are fewer noise pixel points and blurry areas in the spectrogram,so this method can effectively remove noise from the signals and improve the quality of digital audio signals.

关 键 词：生成对抗网络 数字音频 信号多声道 傅里叶变换 双边语谱图 滤波算法 去噪化 全卷积神经网络 

分 类 号：TN911.7-34[电子电信—通信与信息系统] TP391[电子电信—信息与通信工程]