I worked with two team members to create a computer interface that can be controlled with the eye muscle movements of the user. My main role was coding a word recognition program in MATLAB that would serve as an addition to the interface. The program would record a user saying a command word, such as “internet”. This recording would then be saved as a wav file and cross-correlated (take two signals and sweep them across each other) with a library containing the pre-recorded wav files of the commands available. The cross-correlation is used to find out how similar the audio signals are. Two very different signals cross-correlating would result in a low maximum. Two very similar signals cross-correlating would result in a high maximum. The cross-correlated library sound file with the highest maximum would be the matching phrase. The command associated with that file would then execute. Screenshots of the cross-correlation tests can be seen below.
More details on the project can be found on the team blog.
