/ 0
60%
Table of contents

Document in text mode:

IEEETRANSACTIONSONAFFECTIVECOMPUTING1Real-timeRecognitionofAffectiveStatesfromNon-verbalFeaturesofSpeechanditsApplicationforPublicSpeakingSkillAnalysisTomasPfisterandPeterRobinsonAbstract—Thispaperpresentsanewclassificationalgorithmforreal-timeinferenceofaffectfromnon-verbalfeaturesofspeechandappliesittoassessingpublicspeakingskills.Theclassifieridentifiessimultaneouslyoccurringaffectivestatesbyrecognisingcorrelationsbetweenemotionsandover6000functional-featurecombinations.Pairwiseclassifiersareconstructedfor9classesfromtheMindReadingemotioncorpus,yieldinganaveragecross-validationaccuracyof89%forthepairwisemachinesand86%forthefusedmachine.Thepaperalsoshowsanovelapplicationoftheclassifierforassessingpublicspeakingskills,achievinganaveragecross-validationaccuracyof81%andaleave-one-speaker-outclassificationaccuracyof61%.Optimisingsupportvectormachinecoefficientsusinggridparametersearchisshowntoimprovetheaccuracybyupto25%.Theemotionclassifieroutperformspreviousresearchonthesameemotioncorpusandissuc...