Nonlinear Dynamic Feature Extraction Based on Phase Space Reconstruction for the Classification of Speech and Emotion

Table 7

Four types of features used to obtain a confusion matrix for the mixed language emotional speech recognition task.

Feature type	Emotional state	CASIA-Chinese	Berlin-German	Average
Prosody	Happiness	42.65	66.67	54.66
	Sadness	52.94	70.00	64.47
	Neutral	48.53	69.23	58.88
	Anger	69.12	62.96	66.04
	Fear	44.12	17.39	33.76
	Average	51.47	57.50	54.49
MFCC	Happiness	55.88	54.17	55.03
	Sadness	52.94	80.00	66.47
	Neutral	75.00	76.92	75.96
	Anger	69.12	85.19	77.16
	Fear	38.26	34.78	36.52
	Average	58.24	66.67	62.46
NLD-1	Happiness	50.00	54.17	52.09
	Sadness	47.06	60.00	53.53
	Neutral	76.47	80.77	78.62
	Anger	73.53	77.78	76.65
	Fear	44.12	69.57	56.85
	Average	55.29	69.17	62.23
NLD-2	Happiness	52.94	70.83	61.89
	Sadness	48.53	80.00	64.27
	Neutral	44.12	69.23	56.68
	Anger	79.41	85.19	82.30
	Fear	54.11	73.91	64.01
	Average	55.88	75.83	65.86
Prosody + MFCC + NLD	Happiness	75.00	72.06	73.53
	Sadness	76.09	72.06	74.08
	Neutral	77.78	73.53	75.66
	Anger	86.96	79.41	83.19
	Fear	79.17	70.59	74.88
	Average	79.17	73.53	76.35