Recognition of Spanish vowels through imagined speech by using spectral analysis and SVM

Recent works have been studied the possibility of develop a communication system based on EEG signals as a tool for information transmission in people with disabilities. In this work the results of acquiring and analyzing EEG signals related to the communication process in the human being are presented, all this with the aim to identify two vowels of the Spanish language by using imagined speech. In first place, the acquisition of EEG signals through BCI devices was performed, the next step was the treatment of the signals with DSP techniques such as filtering, and the Blackman-Tukey transform, also a dimensionality reduction method known as Symbolic Aggregate Approximation was used for training the Support Vector Machines. The developed algorithm is able to classify and recognize the signals from the thinking of two vowels with an accuracy of 85.29% using six of the fourteen BCI’s sensors that measure the Brodmann areas related with the language process. © 2016.