Evaluation

To evaluate the robustness of the platform, 10 participants (6 female, average age 23.6 years old) were invited in the evaluation experiment. None of them has experienced the same study before, to collect the unseen data. The arithmetic computation application is used as the basis for accuracy evaluation.

For each participant, showed them a total of 750 digits, randomly sequenced on a computer

screen, and instructed the users to read the number to themselves, without producing a sound and moving their lips. The digits were randomly picked from digits 0 to 9, such that each digit exactly appeared 75 times. The data was recorded for each user with the trigger voltages marking word-label alignments.

The experiment was conducted in two phases. First, collect silent speech data from users and evaluated word accuracy on a split training test. Second, assess the recognition latency of the interface by testing live inputs on the model, trained using the previous step.