基于神经网络的Bark域图形均衡器的设计  

Design of Graphic Equalizer of Bark Domain Based on BP Neural Network

作　　者：沈子扬 王明玉 戴海玲 SHEN Ziyang;WANG Mingyu;DAI Hailing(School of New Energy,Nanjing University of Science and Technology,Jiangyin Jiangsu 214443,China;School of Information Engineering,Southeast University,Nanjing Jiangsu 210096,China)

机构地区：[1]南京理工大学新能源学院,江苏江阴214443 [2]东南大学信息科学与工程学院,江苏南京210096

出　　处：《电子器件》2024年第2期536-543,共8页Chinese Journal of Electron Devices

摘　　要：描述了一种基于BP神经网络的图形均衡器设计方法,在不牺牲参数均衡滤波器的Bark频带图形均衡器精度的情况下简化图形均衡器的设计。其核心思想是训练一个神经网络来预测从目标增益到指定中心频率处的优化通带增益的映射关系。在24通道的Bark频带图形均衡器的情况下,利用具有48个神经元隐藏层的BP神经网络的非线性映射功能来实现预测。然后,使用封闭式公式快速、简便地计算频带滤波器的系数。这项工作将引入使用最小二乘法获得滤波器最佳增益的精确控制方法,并不断加以改进。采用BP神经网络与目标增益直接预测获得参数均衡器的优化增益,使得计算量大大减少,使得基于参数均衡滤波器的Bark频带图形均衡器的逼近误差小于0.1 dB。由此产生的BP神经网络控制的24通道Bark域图形均衡器非常有用。A method based on back propagation(BP)neural network is described to simplify the design of graphic equalizer without sacrifi⁃cing approximation accuracy.Its core idea is to train a neural network to predict the mapping relationship between the target gain and the optimized bandpass gain at the specified center frequency.In the case of a 24⁃channel Bark band graphic equalizer,the data fitting func⁃tion of the BP neural network with a hidden layer of 48 neurons is used to realize the prediction.Then,the closed formula is used to calcu⁃late the coefficients of the band filter quickly and easily.The precise control method of using a least square method to obtain the optimal gain of the infinite impulse response(IIR)filter is introduced and continued to be improved.BP neural network and target gain are used to obtain the optimal gain of the parameter equalizer,greatly reducing the amount of calculation and making the approximation error less than 0.1 dB.The resulting neural controlled 24⁃channel Bark domain graphic equalizer is very useful in audio conference equalization requiring time⁃varying equalization.

关 键 词：图形均衡器 BP神经网络 IIR滤波器 

分 类 号：TN912.3[电子电信—通信与信息系统] TP183[电子电信—信息与通信工程]