Skip to main content

Table 9 Test characteristics of previous research using speech analysis and machine learning for AHI classification and regression

From: Reviewing the connection between speech and obstructive sleep apnea

Study

Population characteristics

Classification

Regression

Correct classification rate (%)

Sensitivity (%)

Specificity (%)

Correlation coefficient

GMMs [10]

80 male subjects

(AHI <10: 40 men, AHI >30: 40 men)

81

77.5

85

_

HMMs [11]

80 male subjects

(AHI <10: 40 men, AHI >30: 40 men)

85

_

_

_

Several feature selection and classification schemes [13]

248 subjects

(AHI ≤5: 48 male, 79 women; AHI ≥30: 101 male, 20 women)

82.85

81.49

84.69

_

Feature selection and GMMs [9]

93 subjects

(AHI ≤5: 14 female; AHI >5: 19 female)

(AHI ≤10: 12 male; AHI >10: 48 male)

_

86

83

84

79

_

Feature selection and GMMs [41]

103 male subjects

(AHI ≤10: 25 male; AHI >10: 78 male)

80

80.65

80

_

Feature selection, supervectors and SVR [14]

131 males

_

_

_

0.67a

I-vectors/supervectors and SVR this study

426 males

(AHI <10: 125 male; AHI ≥10: 301 male)

71.06

92.92

20.6

0.30

  1. aResults using speech features plus age and BMI