Brainstorm is a more efficient, noninvasive mind controlled keyboard prototype. In this system, language boxes are flashed to the user in distinct frequency ranges. Selections are made with logistic regression, which classifies box frequency using frequency power bands extracted from the Emotiv EPOC X’s 14 wet electrodes. Brainstorm uses voting to reduce false positives, with each selection required to obtain a majority of votes in a given time bin; on average, this method explained .77 of frequency variance and achieved a classification accuracy of .68 during validation on test data. After each classification, Brainstorm queries google for autocomplete suggestions, which are then presented to the user in a language box. A robust blink detection algorithm based on Z scores pools peaks across electrodes and time, with a series of rapid blinks undoing the last keyboard event. When a user is done typing, the current language output is distributed via an integrated email system. During development, this system was used to send the world’s first recorded email using a noninvasive EEG headset. The entire windows form application is provided as a free download that can be run on any personal computer.
In addition to acting as an end to end system for users, Brainstorm provides developers with an open source eeg data set containing over 50,000,000 labeled data points and source code that includes backtesting, session logging, and data visualization. This combination of open source data and software, will accelerate both the development and adoption of mind controlled keyboards.