Depression-level assessment from multi-lingual conversational speech data using acoustic and text features

EURASIP Journal on Audio, Speech, and Music Processing

Table 5 Best classification results for Turkish Task. Avec 2013 feature set used for all results except Gemaps row. There are 50 depressed and 27 non-depressed subjects in the database. Number of selected features are shown in parenthesis for each case. Results are statistically insignificant except the classification using text-only features

Method	Classes	Precision	Recall	F1-score
*Baseline MRMR(3)*	Non-depressed	0.61	0.40	0.48
	Depressed	0.72	0.86	0.78
	Average	0.67	0.63	0.63
*Tr+Eng N*_min=3 ml-MRMR(40)	Non-depressed	0.58	0.52	0.55
	Depressed	0.75	0.80	0.78
	Average	0.67	0.66	0.66
*Tr+Eng N*_min=5 ml-MRMR(400)	Non-depressed	0.59	0.48	0.53
	Depressed	0.74	0.82	0.78
	Average	0.66	0.65	0.66
*Tr+Ger N*_min=3 ml-MRMR(100)	Non-depressed	0.42	0.29	0.34
	Depressed	0.67	0.78	0.72
	Average	0.54	0.54	0.53
*Tr+Ger N*_min=5 ml-MRMR(15)	Non-depressed	0.47	0.52	0.49
	Depressed	0.72	0.68	0.70
	Average	0.60	0.60	0.60
*Only text features MRMR(7)*	Non-depressed	0.78	0.40	0.53
	Depressed	0.74	0.94	0.83
	Average	0.76	0.67	0.68
*GEMAPS*	Non-depressed	0.38	0.37	0.37
	Depressed	0.66	0.68	0.67
	Average	0.52	0.53	0.52