Non-Invasive Suicide Risk Prediction Through Speech Analysis

TL;DR

This work presents a non-invasive, speech-based approach for automatic suicide risk assessment in emergency medicine, and shows that integrating the model with a series of patients' metadata, such as the history of suicide attempts or access to firearms, improves the overall result.

Abstract

The delayed access to specialized psychiatric assessments and care for patients at risk of suicidal tendencies in emergency departments creates a notable gap in timely intervention, hindering the provision of adequate mental health support during critical situations. To address this, we present a non-invasive, speech-based approach for automatic suicide risk assessment. For our study, we collected a novel speech recording dataset from $20$ patients. We extract three sets of features, including wav2vec, interpretable speech and acoustic features, and deep learning-based spectral representations. We proceed by conducting a binary classification to assess suicide risk in a leave-one-subject-out fashion. Our most effective speech model achieves a balanced accuracy of $66.2\,\%$. Moreover, we show that integrating our speech model with a series of patients' metadata, such as the history of suicide attempts or access to firearms, improves the overall result. The metadata integration yields a balanced accuracy of $94.4\,\%$, marking an absolute improvement of $28.2\,\%$, demonstrating the efficacy of our proposed approaches for automatic suicide risk assessment in emergency medicine.

Figures (1)

• Figure 1: Comparison of models trained solely with metadata (blue) and models with metadata fused with features from different speech recordings: all speech data (green), speech data exclusively from picture descriptions (red), reading neutral texts (purple), and distinct vowels (orange). Feature types F1--F10: cf. \ref{['tab:metadata-fusion']}