This investigation seeks to detect depression via electroencephalography (EEG), comparing the responses of individuals with depression and healthy controls to stimuli involving inner speech (internal dialogue) versus stimuli involving overt speech (spoken words). Participants were recruited in case-control pairs, matched by gender, age, and handedness. Participants completed questionnaires assessing demographics and depression symptoms and then completed the EEG experiment. While wearing an EEG headcap, participants were presented with visual stimuli (words on a screen) and were instructed to read the words first to themselves (“inner speech”) and then out loud (“overt speech”). The word stimuli varied in emotional valence - positive, neutral, and negative. To determine if there were different EEG characteristics between clinical and control subgroups, an EEGNet Machine Learning (ML) model was trained to recognize differences in EEG results between those with depression and healthy controls. The ML model had the best predictive ability for the overt speech cues compared to inner speech and resting state, and specifically, the accuracy of prediction based on cues from this condition was greater than what would be due to chance (50%), regardless of the emotional valence of the word cues. From these findings, the authors reached the conclusion that detection of depression via EEG responses to overt speech is possible using EEGNet technology. The finding that depression can be predicted using EEG data suggests that there is detectable brain activity in specific brain regions associated with depression. One significant limitation is that the predictive ability of the model did not differ statistically from the predictive ability of the Beck Depression Inventory (BDI), which may limit the clinical utility of this finding.