简介：Foraccuracyandrapidityofaudioeventdetectioninthemass-dataaudioprocessingtasks,agenericmethodofrapidlyrecognizingaudioeventbasedon2D-HaaracousticsuperfeaturevectorandAdaBoostisproposed.Firstly,itcombinescertainnumberofcontinuousaudioframestobean'acousticfeatureimage',secondly,usesAdaBoost.MHorfastRandomAdaBoostfeatureselectionalgorithmtoselecthighrepresentative2D-Haarpatterncombinationstoconstructsuperfeaturevectors;thirdly,analyzesthecommonalityanddifferencesbetweensubcategories,thenextractscommonfeaturesandreducesdifferentfeaturestoobtainagenericaudioeventtemplate,whichcansupporttheaccurateidentificationofmultiplesub-classesanddetectandlocatethespecificaudioeventfromtheaudiostreamaccurately.Experimentalresultsshowthattheuseof2D-Haaracousticfeaturesupervectorcanmakerecognitionaccuracy5%higherthanonesthatMFCC,PLP,LPCCandothertraditionalacousticfeaturesyielded,andcanmakethetrainingprocessing7-20timesfasterandtherecognitionprocessing5-10timesfaster,itcanevenachieveanaverageprecisionof93.38%,anaveragerecallof95.03%undertheoptimalparameterconfigurationfoundbygridmethod.Aboveall,itcanprovideanaccurateandfastmass-dataprocessingmethodforaudioeventdetection.