The performance of automated speech recognition systems under adverse conditions of human exertion

Research was conducted to determine if a relation exists between human exertion and the ability of speech recognition software to correctly recognize human speech. Participants were asked to use voice recognition technology to input a short newspaper article in 3 portions. 1 portion of the selected article was read while the participants were rested, another portion while they were lightly exerted, and the final portion while they were experiencing hard exertion. Recognition percentages were computed and compared for rested, lightly exerted, and moderately hard exerted states. The results identified a negative linear relation between physical exertion and recognition accuracy; the higher the level of exertion, the lower the accuracy rate.