Feature Aggregation with Two-Layer Ensemble Framework for Multilingual Speech Emotion Recognition

Table 7

Preclassifiers’ RA results for seven emotions using HDFM as training dataset and Korean KETI dataset as testing dataset.

	Precision	Recall	F1-score
LRC
Anger	0.14	0.08	0.11
Disgust	0.20	0.18	0.19
Fear	0.16	0.33	0.22
Happiness	0.38	0.50	0.43
Neutral	0.17	0.03	0.05
Sadness	0.27	0.38	0.32
Surprise	0.08	0.02	0.03
Accuracy			0.26
Macro Average	0.20	0.22	0.19
Weighted Average	0.23	0.26	0.23
RFC
Anger	0.15	0.63	0.24
Disgust	0.00	0.00	0.00
Fear	0.44	0.07	0.13
Happiness	0.33	0.51	0.40
Neutral	0.18	0.04	0.06
Sadness	0.70	0.05	0.10
Surprise	0.33	0.01	0.01
Accuracy			0.23
Macro Average	0.30	0.19	0.13
Weighted Average	0.30	0.23	0.17
GBC
Anger	0.12	0.11	0.12
Disgust	0.12	0.06	0.08
Fear	0.25	0.38	0.30
Happiness	0.33	0.33	0.33
Neutral	0.14	0.29	0.19
Sadness	0.23	0.15	0.18
Surprise	0.33	0.01	0.01
Accuracy			0.20
Macro Average	0.18	0.20	0.18
Weighted Average	0.20	0.20	0.19
MLP-C
Anger	0.18	0.41	0.25
Disgust	0.23	0.15	0.18
Fear	0.21	0.34	0.26
Happiness	0.36	0.37	0.36
Neutral	0.06	0.00	0.01
Sadness	0.35	0.15	0.21
Surprise	0.16	0.23	0.19
Accuracy			0.25
Macro Average	0.22	0.24	0.21
Weighted Average	0.25	0.25	0.23