Detecting Depression Using an Ensemble Logistic Regression Model Based on Multiple Speech Features

Table 1

Summary of speech features.

Main category	Subcategory	Number of features	Functions
MFCC	MFCC (0–14)	630	Corresponding delta coefficients appended
SPEC	Flux	42	21 functions utilized
	Centroid	42	maxPos, minPos
	Entropy	42	Mean, std dev
	Roll-off	168	Skewness, kurtosis
	Band energies	84	Quartile 1/2/3
PROS	PCM loudness	42	Quartile range (2–1)/(3–2)/(3–1)
	Log mel-frequency band (0–7)	336	Linear regression error Q/A
	LSP frequency (0–7)	336	Linear regression coeff. 1/2
	F0 envelope	42	Percentile 1/99
	Voicing probability	42	Percentile range (99–1)
	F0final, ShimmerLocal	76	19 functions by eliminating the minimum value and the
	JitterLocal, JitterDDP	76	Range functions from the 21 abovementioned functions
	Pitch onsets, duration	2	No functions
GLOT	GLT	27	Mean, max, min
	GLF	5	Mean, max, min
Total		1992