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.
CITATION STYLE
Baykaner, K., Huckvale, M., Whiteley, I., Ryumin, O., & Andreeva, S. (2015). The prediction of fatigue using speech as a biosignal. In Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) (Vol. 9449, pp. 8–17). Springer Verlag. https://doi.org/10.1007/978-3-319-25789-1_2
Mendeley helps you to discover research relevant for your work.