The Prediction of Fatigue Using Speech as a Biosignal
- Cite this paper as:
- Baykaner K., Huckvale M., Whiteley I., Ryumin O., Andreeva S. (2015) The Prediction of Fatigue Using Speech as a Biosignal. In: Dediu AH., Martín-Vide C., Vicsi K. (eds) Statistical Language and Speech Processing. Lecture Notes in Computer Science, vol 9449. Springer, Cham
Automatic systems for estimating operator fatigue have application in safety-critical environments. We develop and evaluate a system to detect fatigue from speech recordings collected from speakers kept awake over a 60-hour period. A binary classification system (fatigued/not-fatigued) based on time spent awake showed good discrimination, with 80 % unweighted accuracy using raw features, and 90 % with speaker-normalized features. We describe the data collection, feature analysis, machine learning and cross-validation used in the study. Results are promising for real-world applications in domains such as aerospace, transportation and mining where operators are in regular verbal communication as part of their normal working activities.