Wearable Stress Detector Hits 94% Accuracy by Reading Your Voice and Sweat
A wearable wristband combined with a badge that listens to voice and motion pushes binary stress detection to 94 percent accuracy during a lab stress test.
Takeaways
- Combined wristband and sociometric badge hits 94% accuracy on binary stress detection.
- Tested on 18 participants using the Trier Social Stress Test with public speaking and arithmetic.
- AdaBoost with 300 weak learners beat RBF SVM, linear SVM, and KNN.
- Sociometric badge alone reached 89%; physiological wristband alone only 79%.
- Personalized classifiers required; a cross-subject general model dropped to 67%.
- Top features: EDA, posture activity, mic frequency peaks, and body movement.
Detecting when someone is stressed usually means strapping sensors to their body and hoping a heart rate spike gives them away. Researchers from the Technical University of Cartagena, the University of Lincoln, and collaborators tried a different angle: pair a physiological wristband with a sociometric badge that captures voice and movement, then train a personal classifier for each subject. The combined system hits 94% accuracy discriminating stressful from neutral situations during a Trier Social Stress Test, and the paper spells out which sensor modality actually carries the signal.
Why body signals alone keep falling short
Stress shows up through variations in heart rate, pulse, skin temperature, pupil dilation, and electro-dermal activity, and prior work has tried to read it from each of these individually. One catch keeps tripping earlier systems: responses vary wildly between people, and lab setups often generate stress through computer tasks like the Stroop test rather than realistic social pressure. Pupil-diameter systems, for example, work in the lab but are hard to wear day to day.