Improved Emotion Recognition Using Gaussian Mixture Model and Extreme Learning Machine in Speech and Glottal Signals

Table 3

Average emotion recognition rates for BES database.

Different order “db” wavelets	Speaker dependent
	Raw features	Enhanced features	SET1	SET2	SET3	SET4
ELM kernel
db3	69.99	98.24	98.52	98.15	97.50	98.15
db6	68.55	95.65	96.39	97.31	96.67	97.13
db10	68.77	98.52	97.78	97.69	96.85	96.39
db44	67.43	98.70	98.43	98.98	98.89	98.61
kNN
db3	59.14	95.19	96.57	96.11	95.83	96.30
db6	58.68	89.72	87.31	91.30	90.65	91.67
db10	57.93	97.04	96.57	96.57	95.46	95.28
db44	57.63	95.46	95.56	95.56	95.00	96.11
Speaker independent
ELM kernel
db3	56.61	96.04	97.07	97.24	96.40	96.40
db6	54.70	92.26	91.26	91.83	91.62	91.48
db10	52.08	92.12	94.13	93.37	93.00	92.93
db44	53.60	93.04	93.13	94.10	93.67	94.33
kNN
db3	49.12	91.75	93.32	92.01	92.79	93.90
db6	48.21	81.93	82.22	81.43	82.18	82.77
db10	45.17	90.64	89.81	89.46	90.23	90.15
db44	46.87	91.69	90.15	90.57	90.35	91.99