Word Prediction Support Model for SSVEP-Based BCI Web Speller

Steady state visual evoked potential (SSVEP) based BCI-systems are dependent on the brain signals which are elicited in response to a visual stimuli presented to the user. The spelling systems are very popular applications for the SSVEP-based BCI. In this paper, we are presenting a web-based speller supported with word prediction. The emphasis of the study was on two main points: (1) provide a dictionary based web speller which could also be accessed through a widely available web browsers; (2) increase the accuracy and speed of the SSVEP-based BCI speller. Using the concept of three step speller, a web interface was developed which provided additional support of word predictions based on characters typed and the co-occurrence of the previously typed word. The architectural pattern for the word prediction support model was based on MVC (model-view-controller). The AJAX call was placed form the web speller interface to access the database using Java Servlet and Java Beans. The relational database for the word prediction was derived from the Leipzig corpora collection. The developed system was tested with eleven healthy subjects. An average accuracy of 92.5 % and ITR of 18.8 bits/min were achieved. The results showed that word suggestions can increase the typing speed and accuracy of the web speller.