Evaluating automatic speech recognition as a component of a multi-input device human-computer interface

This paper reports on an investigation into the basic properties and requirements of automatic speech recognition as an input device to a trial human computer interface. The experiments required subjects to carry out a simulated target acquisition and report filling task, with the available input devices being automatic speech recognition, trackball, function keys or a simultaneous combination of all three. Experiments were carried out under varying workload to examine the degradation of overall interface and individual input device performance under user stress. An approach at modelling interface performance using a critical path analysis approach is introduced. Modelling of the interface developed here has shown a good match to the experimental results. Although use of the prototype speech recogniser was found to be both slower and less accurate than the manual mode inputs it was possible to estimate a required word accuracy of around 94% which would allow speech entry to provide an equivalent performance.