基于深度学习的光纤麦克风频带扩展  

The fiber optic microphone bandwidth expansion based on deep learning

作　　者：方健[1,2] 甄胜来[1,2] 陈鑫[1,2] 俞本立[1,2] FANG Jian;ZHEN Shenglai;CHEN Xin;YU Benli(Key Laboratory of Opto-electronic Information Acquisition and Manipulation,Ministry of Education,Anhui University,Hefei 230601,China;Information Materials and Intelligent Sensing Laboratory of Anhui Province,Anhui University,Hefei 230601,China)

机构地区：[1]安徽大学光电信息获取与控制教育部重点实验室,安徽合肥230601 [2]安徽大学信息材料与智能感知安徽省实验室,安徽合肥230601

出　　处：《安徽大学学报（自然科学版）》2024年第3期39-45,共7页Journal of Anhui University(Natural Science Edition)

基　　金：安徽省重点研究与开发计划项目(202104a05020059)

摘　　要：光纤麦克风具有体积小、精度高、抗干扰能力强等优点,能在复杂环境下拾取目标语音.然而,在采集语音过程中,光纤麦克风受响应带宽限制,出现了高频成分缺失情况,进而降低语音短时客观可懂度(short-time objective intelligibility,简称STOI)和信噪比(signal-to-noise ratio,简称SNR).将时间卷积模块(temporal convolutional module,简称TCM)引入Wave-U-Net,提出TCM_Wave-U-Net.在此基础上,提出频域卷积递归神经网络(convolutional recurrent neural networks,简称CRN)与时域TCM_Wave-U-Net协同的网络(简称协同网络).实验结果表明:协同网络具有较强的泛化性和鲁棒性.该文研究结果为光纤麦克风的语音保真拾取奠定了基础.The optical fiber microphone has the advantages of small size,high precision and strong anti-interference ability and can pick up the target speech in a complex environment.However,in the process of speech acquisition,the optical fiber microphone is limited by the response bandwidth,and high-frequency components are missing,which reduces the short-time objective intelligibility(STOI)and signal-to-noise ratio(SNR)of speech.The temporal convolutional module(TCM)was introduced into Wave-U-Net,and TCM_Wave-U-Net was proposed.On this basis,the collaborative network between frequency-domain convolutional recurrent neural networks(CRN)and time-domain TCM_Wave-U-Net(It was called collaborative network for short)was proposed.The experimental results showed that the cooperative network had strong generalization and robustness.The research results of this paper laid a foundation for the speech fidelity pickup of the optical fiber microphone.

关 键 词：光纤麦克风 频带扩展 深度学习 卷积神经网络 多尺度融合 

分 类 号：TN29[电子电信—物理电子学] TP391[自动化与计算机技术—计算机应用技术]