Detecting Depression Using an Ensemble Logistic Regression Model Based on Multiple Speech Features
Algorithm 1
ELRDD.
Input: training speech recordings and its label and testing speech recordings .
Output: depressed patient or healthy control labels of .
//Training process
Step 1: extract MFCC, PROS, SPEC, and GLOT features for each speech recording from , and compute the feature statistics as listed in Table 1.
Step 2: in terms of Table 2, 15 feature subspaces are constructed , where .
For k = 1 to 15
Step 3: feature reduction for X(k) is achieved using LLE.
End
Step 4: maximize Equation (3) to achieve the trained classifier model.
//Testing process
Step 5: extract MFCC, PROS, SPEC, and GLOT features for each speech recording from , and compute the feature statistics as listed in Table 1.
Step 6: in terms of Table 2, 15 feature subspaces are constructed , where .
For k = 1 to 15
Step 7: feature reduction for D(k) is achieved using LLE.
End
Step 8: based on the trained classifier model, apply Equations (1) and (2) to compute the probabilities that the testing samples belong to depressed patients or healthy controls , and then output the category label whose probability is greater.