Abstract
Speech rhythm metrics were explored as a potential informative biomarker for the classification of depressed speech. A database of speech recordings collected from depressed patients undergoing treatment was automatically segmented via text alignment processes. Speech rhythm metrics were then calculated for each of the speech samples with a script written specifically for this analysis. Rhythm metric values were correlated with depression severity scores to determine the level of co-variation. Silence was found to increase and consonant intervals to display less variability with increased depression. Results thus indicate a slower speaking rate and a form of consonant reduction in depressed speech. However, no metrics were consistently correlated with depression across all elicitation tasks, depression evaluation scales, or speakers, suggesting further investigation into optimal condition-specific features is merited.