JMIR Publications

ABSTRACT

Background:

The stress burden associated with family caregiving can make family caregivers particularly susceptible to psychosocial health conditions. However, early detection and intervention can help prevent disease progression and long-term disability.

Objective:

This study developed the Linguistic and Acoustic Speech Analytics Program (LASAP) with the aim of providing a comprehensive fusion analysis of both linguistic and acoustic features of speech, in order to enhance psychosocial stress assessment.

Methods:

A quantitative study that analyzed speech data collected from 100 Chinese family caregivers employed various machine learning classifiers to evaluate psychosocial stress levels. Speech data were processed using machine learning models, including the Linear Support Vector Machine (SVM). The analysis involved a fusion of linguistic and acoustic features, and an orthogonalization procedure to decorrelate features before fusion. Model performance was measured using ROC-AUC and accuracy metrics.

Results:

The Linear SVM achieved an ROC-AUC score of 78.28% and two accuracy-related scores of 75.27% and 73%. The fusion analysis of linguistic and acoustic features significantly improved classification power compared to using either feature type alone. Furthermore, the orthogonalization procedure, which decorrelated acoustic features from linguistic features before fusion analysis, markedly enhanced classification accuracy.

Conclusions:

This study demonstrates that integrating linguistic and acoustic feature analyses effectively identifies psychosocial stress in family caregivers. It also emphasizes the importance of proper feature processing when combining multiple features extracted from the same audio sample. These findings provide valuable insights for developing machine learning models for psychosocial stress assessment and addressing various psychosocial conditions in different contexts, supporting population mental health management.