Automated Vocal Fatigue Screening in Professional Voice Users: Development and Occupational Validation of an Automated Assessment System.

INTRODUCTION Professional voice users face an elevated risk of vocal injury from sustained occupational voice use, yet scalable tools for workplace surveillance remain limited.

Objective

To develop and validate an automated vocal fatigue monitoring system for occupational health applications.

Methods

We developed a speaker-independent assessment system using deep learning embeddings trained on diverse speech samples. Validation included a correlation with the Vocal Fatigue Index (VFI) in 27 participants (9 teachers, 18 general population) and temporal sensitivity assessment using twice-daily recordings from 8 teachers over 5 workdays.

Results

Automated scores demonstrated very strong correlation with VFI (r = 0.942, P < 0.001), increased significantly across work shifts (d = 1.00, P < 0.001), and correlated with speaking duration (r = 0.544, P = 0.0003). Teachers showed substantially higher fatigue than the general population (d = 3.21, P < 0.001).

Conclusions

The system demonstrates promising early validation for automated shift-level screening and speaker-independent assessment of occupational vocal load for integration into workplace health surveillance and vocal injury prevention programs.