<?xml version="1.1" encoding="utf-8"?>
<article xsi:noNamespaceSchemaLocation="http://jats.nlm.nih.gov/publishing/1.1/xsd/JATS-journalpublishing1-mathml3.xsd" dtd-version="1.1" xmlns:xlink="http://www.w3.org/1999/xlink" xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"><front><journal-meta><journal-id journal-id-type="publisher-id">TACS</journal-id><journal-title-group><journal-title>Technology and Application of Computer Science</journal-title></journal-title-group><issn>2998-8926</issn><eissn>2998-8934</eissn><publisher><publisher-name>Art and Design</publisher-name></publisher></journal-meta><article-meta><article-id pub-id-type="doi">10.61369/TACS.2025080019</article-id><article-categories><subj-group subj-group-type="heading"><subject>Article</subject></subj-group></article-categories><title>基于Conformer 的咳嗽声检测</title><url>https://artdesignp.com/journal/TACS/2/8/10.61369/TACS.2025080019</url><author>乔凯,李哲宏,姚袁,徐金诚,王明欢</author><pub-date pub-type="publication-year"><year>2025</year></pub-date><volume>2</volume><issue>8</issue><history><date date-type="pub"><published-time>2025-04-28</published-time></date></history><abstract>通过声音信号检测咳嗽对于医疗和健康监护的应用至关重要，包括疾病诊断和远程患者监测。虽然很多深度学习算法在实验数据上能够取得90% 以上的准确率，但是一旦应用于实际环境中，咳嗽检测的准确率就会大大降低。很多种类的声音（例如：短促的说话声、笑声、关门声等）都被识别为咳嗽。卷积神经网络（CNN）联合循环神经网络（RNN）虽能能够提升检测性能，但在捕捉咳嗽各个阶段的依赖关系和时序动态特征方面仍面临挑战。为此，本文提出用于咳嗽检测的Cough-Conformer（卷积增强型Transformer）架构，通过卷积层实现局部特征提取，并结合自注意力机制进行全局上下文建模。我们在咳嗽数据集COSWARA、语音数据集、噪声数据集、和笑声数据集的基础上提取声音数据作为实验数据集；在该数据集上训练Cough-Conformer，准确率达到了97.64%，F1得分为 0.98，然后在计算机房录制的音频数据集上验证，其中准确率达到了87.01%，F1得分为 0.87。实验结果表明，Cough-Conformer 在咳嗽检测任务中相比于传统CNN 和RNN 模型，在准确率和F1得分均有显著提升，尤其在复杂背景噪声环境下表现出更强的健壮性。通过引入多头自注意力机制，模型能够有效捕捉咳嗽声音的时序动态特征与上下文依赖关系。进一步分析显示，卷积层与Transformer 模块的协同作用提升了对咳嗽不同阶段特征的辨识能力，为远程患者监测中的咳嗽检测提供了更优秀的解决方案。</abstract><keywords>咳嗽声检测,Cough-Conformer 模型,时序数据</keywords></article-meta></front><body/><back><ref-list><ref id="B1" content-type="article"><label>1</label><element-citation publication-type="journal"><p>[1]Zhang P C , Wang Y H , Liu X ,et al.Conformational study of 8-C-glucosyl-prunetin by dynamic NMR spectroscopy[J].Chinese Chemical Letters, 2002, 13(7):645-648.DOI:10.1021/cm020249a.[2] Li-Xin Y .Conformation Analysis and Comparison of Epristeride(17&amp;beta;-N-t-Butylcarboxamide-androst-3,5-diene-3-carboxylic Acid) and Its Analogs[J].高等学校化学研究：英文版, 2005, 21(5):3.DOI:CNKI:SUN:GHYJ.0.2005-05-005.[3]ZHANG, Wang P C , Liu Y H ,et al.Conformational Study on 8-C-glucosyl-prunetin by Dynamic NMR Spectroscopy[J].Acta Chimica Sinica, 2003.[4] 俞涵. 中英文混合的民航空管语音识别研究[D]. 厦门大学,2022.[5] Sun Y , Zhang F , Zhang L ,et al.Synthesis of calix[4]arene derivatives via a Pd-catalyzed Sonogashira reaction and their recognition properties towards phenols[J]. 中国化学快报( 英文版), 2014.[6] Zhu Y B .STEREOCHEMICAL CONTROL IN PROPYLENE POLYMERIZATION CATALYZED BY UNBRIDGED METALLOCENE CATALYSTS[J]. 高分子科学( 英文版),2001.[7]Sun Y , Pan W , Fu J ,et al.Conformation preference and related intramolecular noncovalent interaction of selected short chain chlorinated paraffins[J]. 中国科学: 化学( 英文版), 2016.[8] Zge B , Grkan K , Cemal P ,et al.Vibrational investigation of 1-cyclopentylpiperazine:A combined experimental and theoretical study[J]. 中国科学: 物理学 力学 天文学( 英文版), 2014.[9]STUDIES ON THE CONFORMATIONS OF SUBEROGORGIN AND ITS METHYL ESTER BY MNDO METHOD[J]. 科学通报： 英文版, 1990(23):4.DOI:CNKI:SUN:JXTW.0.1990-23-016.[10]Renqing L , Zuogang C , Guoping S .Ab Initio Calculation of Room Temperature Ionic Liquid 1-Ethyl-3-Methyl-Imidazolium and AlCl3[J].China Petroleum Processing &amp;amp; Petrochemical Technology, 2007, 16(3):51-56.DOI:10.1007/s10553-007-0078-7.[11] Hongwei K E , Li R , Amp X X .Density functional theory study of 1:1 glycine&amp;ndash;water complexes in the gas phase and in solution[J]. 中国科学: 化学( 英文版), 2010.[12]Subhasish,Bandyopadhyay,Asit,et al.Intra-species sequence variability in 28s rRNA gene of Oesophagostomum venulosum isolated from goats of West Bengal,India.[J]. 亚太热带医药杂志: 英文版, 2010(7):515-515.</p><pub-id pub-id-type="doi"/></element-citation></ref></ref-list></back></article>
