Evaluates combined acoustic and linguistic features for detecting depression in Spanish speakers, suggesting improved detection methods.
Key Points
This study aims to assess the role of linguistic markers alongside acoustic features in identifying depression in a clinical, Spanish-speaking population.
Evaluated 151 participants, including 80 with major or persistent depressive disorder and 71 healthy controls.
Participants answered 11 open-ended questions about depressive symptoms via a web platform.
Extracted linguistic and acoustic variables across prosodic, cepstral, spectral, and TEO-based categories.
Performed group comparisons and logistic regressions to analyze predictive values of features.
Utilized machine learning models to compare acoustic, linguistic, and ensemble classification performance.
TEO-based and cepstral features demonstrated the strongest predictive power for depression.
Linguistic features such as verb usage and vocabulary size were strong depression predictors after covariate adjustments.
The linguistic model outperformed the acoustic model with an AUC of 0.86 compared to 0.79.
The ensemble model combining both features achieved an accuracy of 0.84 and specificity of 0.93.
Optimal performance for individuals aged ≤45 years was noted, with an AUC of 0.90.