Speech emotion recognition is currently an\r\nactiveresearchsubjectandhasattractedextensiveinterest\r\nin the science community due to its vital application to\r\nhumanrobot interaction. Most speech emotion\r\nrecognition systems employ highdimensional speech\r\nfeatures, indicating human emotion expression, to\r\nimproveemotionrecognitionperformance.Toeffectively\r\nreduce the size of speech features, in this paper, a new\r\nnonlinear dimensionality reduction method, called\r\nenhanced kernel isometric mapping (EKIsomap), is\r\nproposedandapplied forspeechemotionrecognition in\r\nhumanrobotinteraction.Theproposedmethodisusedto\r\nnonlinearly extract the lowdimensional discriminating\r\nembedded data representations from the original high\r\ndimensionalspeechfeatureswithastrikingimprovement\r\nofperformanceonthespeechemotionrecognitiontasks.\r\nExperimental results on the popular Berlin emotional\r\nspeech corpus demonstrate the effectiveness of the\r\nproposedmethod.
Loading....