Fatigue estimation using voice analysis

In the present article, we present a means to remotely and transparently estimate an individual's level of fatigue by quantifying changes in his or her voice characteristics. Using Voice analysis to estimate fatigue is unique from established cognitive measures in a number of ways: (1) speaking is a natural activity requiring no initial training or learning curve, (2) voice recording is a unobtrusive operation allowing the speakers to go about their normal work activities, (3) using telecommunication infrastructure (radio, telephone, etc.) a diffuse set of remote populations can be monitored at a central location, and (4) often, previously recorded voice data are available for post hoc analysis. By quantifying changes in the mathematical coefficients that describe the human speech production process, we were able to demonstrate that for speech sounds requiring a large average air flow, a speaker's voice changes in synchrony with both direct measures of fatigue and with changes predicted by the length of time awake.