欢迎来到天天文库
浏览记录
ID:36422193
大小:6.85 MB
页数:114页
时间:2019-05-10
《基于内容的音频信息分类检索技术研究》由会员上传分享,免费在线阅读,更多相关内容在学术论文-天天文库。
1、南京理工大学博士学位论文基于内容的音频信息分类检索技术研究姓名:何新申请学位级别:博士专业:控制科学与工程指导教师:周献中20070701摘要博士论文议的录音中包含着大量的语音信息。由于语音具有直观、自然、方便使用的特点,因此,如何直接利用语音来有效地检索多媒体中相关的音频信息,也是一个值得探讨的研究方向。本文针对广播新闻,进行音频检索探讨,研究其中的音频分类、音频检索以及语音识别问题。并在此基础上,设计和初步实现一个基于内容的音频信息检索原型系统。关键词:基于内容的音频信息检索,音频特征,语音识别,模式识别,独立
2、分量分析,支持向量机,广义模型,n博士论文基于内容的音频信息分类检索技术研究AbstractRapidadvancesinthemultimediatechnologyandcomputerprocessingcapacitymakepeoplebefacedwithlargedigital“InformationOcean”.It’SmoreandmoreimportantforpeopletOretrievaltheseinformationquicklyandeffectively.Therefore,mul
3、timediainformationretrievalhasbeenrapidlydevelopedfromthe1990sandbecomeoneofthemostimportantareasintheinformationretrieval.Andinthebeginningmostworkswerefocusedoncontent·basedimageretrievalandcontent—basedvideoretrieval.However,researchesonaudioretrievalarerel
4、ativelagging,justbecauseabundantsemanticinformationcontainedinaudiodataisalwaysignored,andaudiodataarenon—structuring.Asmoreandmoreaudiodataappear,contentbasedaudioretrievalhasbeenoneofresearchhotspotsinmultimediaretrieval.Thisdissertation,whichiSbasedont}lesu
5、mmarizationofformerresearchfindings,dealswithseveralproblemsofcontent—basedaudioretrieval.Researchesareemphasizedonaudiofeatureanalysis,classifierdesignandspeechinformationretrieval.Themainresearchcontentsandresultsofthisdissertationcanbeconcludedasfollows:(1)
6、ResearchOilaudiofeatureclassificationIt’Salwaysbasedonsubjectiveorobjectiveaudiofeaturesforaudioclassification,andtheselectionofaudiofeaturesmustberepresentedimportantclassificationfeaturesintimedomainandfrequencydomain.Hence,analysesandextractionsofaudiofeatu
7、resarethebaseofaudioclassification.HowtOextractfeaturesandkeepthemindependentmutuallyistheimportantproblemstobesettled,whichcailreduceinformationredundancy.Inthispaper,theIndependentComponentAnalysis(ICA)methodisintroducedintoaudiofeatureanalysis.Themethodcail
8、extractpivotalandhighdimensionindependentfeaturesandimprovefeatureseparability.Furthermore,theSupportVectorMachine(SVM)isusedwithitsapprovedclassifiedperformancetOclassifyaudiodata
此文档下载收益归作者所有