The ability to monitor cognitive load level in real time is extremely useful for preventing fatal operating errors or improving the efficiency of task execution. In top of the success of our previously proposed speech based cognitive load monitoring system, we explored alternative classification techniques in this paper, including simple linear kernel Support Vector Machine (SVM), hybrid SVM-GMM which accepts the likelihood scores from GMM as inputs for SVM, and a fusion approach which integrates GMM, SVM and SVM-GMM systems together. All systems are evaluated on the data collected from two different tasks - a reading comprehension and a Stroop test based task. SVM-GMM based system achieved the highest performance on both tasks and improved the accuracy of three cognitive load levels classification from 71.1% to 75.6% and 77.5% to 82.2%, respectively.
Bibliographic reference. Yin, Bo / Ruiz, Natalie / Chen, Fang / Ambikairajah, Eliathamby (2008): "Exploring classification techniques in speech based cognitive load monitoring", In INTERSPEECH-2008, 2478-2481.