欢迎来到天天文库
浏览记录
ID:36423231
大小:7.24 MB
页数:61页
时间:2019-05-10
《基于蚁群算法的说话人识别系统的研究》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、北京交通大学硕士学位论文基于蚁群算法的说话人识别系统的研究姓名:钟明英申请学位级别:硕士专业:模式识别与智能系统指导教师:刘湘黔20080501j丝塞变塑太堂亟±堂缱途塞生塞擅噩说话人识别系统有着较高的识别率,是说话人识别的一种有效的新方法。关键词:说话人识别;特征提取;美尔倒谱系数;混合高斯模型;聚类;蚁群算法分类号:TP391jE瘟銮适塞堂亟±堂缱i佥塞△旦墨!B△£!ABSTRACTABSTRACT:Speakerrecognitiondealswithrecognizingtheidentityofthepersonspeakingutterance;it
2、istheprocessofautomaticallyrecognizingwhoisspeakingbasedontheinformationobtainedfromthespeech.Speakerrecognitionhasawiderangeofapplicationswhichincludebankingorcreditcardtransitionsbytelephone,informationandreservationservices,accesscontrolinhi曲securityareasandforensicinvestigations.T
3、houghspeakerrecognitionsystemperformwellwhencleanspeechisusedfortrainingandtesting,theperformancedegradesrapidlywhenspeechusedinreal—worldconditions.ThefocusofthisresearcheffortistodeveloptechniquesfortrainingGMMparameters.AntColonyOptimization(ACO)isarandomsearchalgorithm.Similartoth
4、eotherevolutionaryalgorithms,itfindstheoptimalsolutionbasedontheevolutionaryprocessofcandidategroupofsearchingpaths.ACOspeedsuptheevolutionaryprocessbypositivefeedbackofpheromone.Itisalsoaparallelalgorithm.Throughcommunicationandtransferringamongtheants,theycouldcollaboratewitheachoth
5、erforfindingthebestpath.ACOiswidelyappliedonTravelingSalesmanProblem(TSP)andotheroptimizationproblems.AfteranalyzingthecurrentresearchesonspeakerrecognitionandAntColonyOptimization(ACO)algorithm,inthisthesis,introducingtheACOtotheparametertrainingprocessofGaussianmixturemodelforbuildi
6、ngspeakerrecognitionsystemWasproposed.AlsotheTopNmethodforrecognitioncomparingbasedoneigenvectorsWasproposed.Aspeakerrecognitionsystemwasrealizedbasedonourapproaches.Themaincontributionsinthethesisareasfollows:1.Thecurrentresearchworksonspeakerrecognitionsystemanditsbasicstructureandt
7、heassessmentcriteriaofrecognitioncorrectnessWasdiscussed.2.TheACOalgorithmWaSanalyzedandthetheoriesandcharactersofACOandGaussianmixturemodelwerestudied.ThenintroducingACOtotheparametertrainingprocessofGaussianmixturemodelWasproposed.Giventhespeechsignalofspeakers,themodelparameterswer
8、eaSse
此文档下载收益归作者所有