Abstract:

Prime III is a multimodal voting system that allows users to use touch or voice to make selections on their ballot. This paper discusses an experiment that evaluated the system’s speech recognition at various levels of background noise. An approach to simulate realistic background noise in a controlled environment is described. This approach helped mimic a voter voting in a precinct. The goal of the experiment was to establish a threshold for when distortion occurs and speech recognition accuracy declines. The signal-to-noise ratios (SNR) between the volumes were recorded and the system’s accuracy was tested. The result was a suggested threshold of a SNR equal to 1.44 to attain 90% system accuracy. The next phase of this project is to test the level of system interference from ambient noise in an actual voting precinct.