Reviewing the connection between speech and obstructive sleep apnea

Table 2 Implementation tools

Tool^a	Function name	Function description	Parameters
HTK	HCopy	Extract the MFCCs coefficients	No. DFT bins = 512
			No. filters = 26
			No. MFCC coeff. = 19
			No. ΔMFCC coeff. = 19
MSR Identity ToolBox^b	GMM_em	GMM–UBM training	No. mixtures = 512
			No. of expectation maximization iteration = 10
			Feature sub-sampling factor = 1
	MapAdapt	GMM adaptation	Adaptation algorithm = MAP
			No. mixtures = 512
			MAP relevance factor = 10
	Train_tv_space	Total variability matrix training	Dimension of total variability matrix = {400,300,200,100,50,30}
	Train_tv_space	Total variability matrix training	Number of iteration = 5
	Extract_ivector	I-vector training	Dimension of total variability matrix = {400,300,200,100,50,30}
LIBSVM	SVM_train	SVR training	Grid search parameters: C, model complexity = −20:20 \(\in\), insensitive-zone = 2⁻⁷:2⁷
LIBSVM	SVM_predict	SVR regression	Grid search parameters: C, model complexity = −20:20 \(\in\), insensitive-zone = 2⁻⁷:2⁷

^aAll the implementation tools were used under Linux Ubuntu 12.04 LTS Operating System
^bExecuted on Matlab 2014a

ISSN: 1475-925X