Speech Feature Identification Model for Depressed Individuals with Suicidal Ideation Based on Autobiographical Memory

Abstract Background: Suicidal ideation in depression is a critical predictor of suicide risk, yet its early identification remains a significant challenge. This study aimed to explore the integration of vocal features and autobiographical memory to develop an objective tool for predicting suicidal ideation in patients with depression. Methods: This study enrolled 88 patients with depression, who were divided into three groups: a group with mild depression and no suicidal ideation, a group with moderate depression and suicidal ideation, and a group with severe depression and suicidal ideation. Methodologies included the Autobiographical Memory Test, various depression scales, and vocal feature extraction techniques. We used statistical methods such as repeated measures analysis of variance (repeated measures ANOVA) and machine learning models to evaluate the differences between groups and to predict suicidal ideation. Results: Significant differences were found in the specificity of autobiographical memory across the groups; patients without suicidal ideation recalled more specific memories than those with suicidal ideation. Regarding vocal features, individuals with suicidal ideation exhibited more monotonous intonation and less pitch variation, as confirmed by metrics such as Mel-frequency cepstral coefficients (MFCCs). Ultimately, a Random Forest model that utilized these features demonstrated high accuracy in the classification task, effectively distinguishing between patients with and without suicidal ideation. Conclusion: The integrated analytical approach of using vocal features and autobiographical memory shows great potential as an objective tool for predicting suicidal ideation. This method provides a valuable tool for the early clinical identification of suicide risk and may lead to more timely and effective interventions.