现代电子技术2025，Vol.48Issue(10)：20-24,5.DOI:10.16652/j.issn.1004-373x.2025.10.004
基于LeNet-RES的室内声源区域定位算法
Indoor sound source regional localization algorithm based on LeNet-RES
摘要
Abstract
Under the conditions of low signal-to-noise ratio(SNR)and high reverberation,indoor sound sourceregional localization becomes more challenging.In order to solve this problem,a neural network LeNet-RES is designed,which uses residual blocks to improve LeNet,thereby improving the performance of the network.The dataset for indoor sound source is obtained by simulating the room impulse response of an 8-array cuboid microphone array.The signal received by the microphone is processed into frames,and the generalized cross correlation PHAse transformation(GCC-PHAT)between each frame signals is calculated.This function is arranged into two-dimensional data as input features.The final network model is trained by taking the room partition area label as the network output.In the experiment,the positioning accuracy of the two neural networks was tested when the number of room partitions was 8 and 16,respectively.The results show that under the same SNR conditions,the accuracy of LeNet-RES-16 is 81.33%when the number of room partitions is 16 and the reverberation time is 0.6 s,which is 23%higher than that of LeNet-16;under the same reverberation conditions,the accuracy of LeNet-RES-16 is 84.16%when the number of partitions is 16 and the SNR is 0,which is 29%higher than that of LeNet-16.The regional localization performance of LeNet-RES is better than that of LeNet under various SNR and various reverberation times.关键词
室内声源定位/麦克风阵列/神经网络/相位变换加权广义互相关函数/信号处理/房间脉冲响应Key words
indoor sound source localization/microphone array/neural network/generalized cross correlation PHAse transformation/signal processing/room impulse response分类
电子信息工程引用本文复制引用
延浩浩,杨瑞峰,郭晨霞..基于LeNet-RES的室内声源区域定位算法[J].现代电子技术,2025,48(10):20-24,5.基金项目
山西省中央引导地方科技发展自由探索类基础研究项目(YDZJSX2022A027) （YDZJSX2022A027）
