Skip to main content

Table 7 Emotion classification results from three-dimensional emotion primitives

From: Recognizing emotion from Turkish speech using acoustic features

 

Neutral

Sad

Happy

Angry

Neutral

Sad

Happy

Angry

Neutral

1,955

53

57

85

1,430

24

0

696

Sad

114

342

1

110

371

11

0

185

Happy

93

0

302

9

262

6

1

135

Angry

163

50

2

1,162

947

13

0

417

WA recall

83.6%

41.3%

UA recall

77.6%

24.8%

  1. Left confusion matrix: emotion classification from the ground truth emotion primitives. Right confusion matrix: emotion classification from the estimates of emotion primitives. SVR is used to estimate the three emotion primitives. For classification, BayesNet classifier is used. The values in italics of the confusion matrices represent the number of correctly predicted emotion labels.