Silent speech classification based upon various feature extraction methods

Envisioned speech captured by EEG signals is a fascinating area of research as this is useful in bio-medical applications for the patients suffering from motor neuron diseases and also in those areas where silent speech is desirable. This work examines the possibility of better feature extraction techniques from which a robust model with the help of a classifier can be built. Results showed that Common Spatial Patterns (CSP) filter coefficients in a combination of statistical features with Random Forest as a classifier turn out to be a suitable choice. Three imagined tasks /a/, /u/ and /rest/ were discriminated with highest accuracy reaching up to 89%.