(13) Silent speech BCI: Learning and decoding algorithms using single-trial EEGs and speech signals
K. Yamamoto, T. Yamazaki, K. Kamijo, T. Yamanoi, S. Fukuzumi
第26回生体・生理工学シンポジウム論文集，pp.623-625，2011-9
This study processes a new silent speech BCI using both single-trial EEGs and speech signals. The EEGs segments, corresponding to the NS' components, are extracted from 19ch single-trial EEGs recorded during the janken task where the subject spoke in Japanese. After analyzing the EEG by ICA, a Kalman filter was learned both from the reconstructed EEGs, whose dipole sources were localized to the premotor cortex by ECDL, and from spectrograms of the speech signals simultaneously recorded. The model spectrograms were predicted from single-trial EEGs recorded during the silent janken tasks. The present results were that the silent "rock" and "paper" ICs were correctly recognized in terms of that the spectrograms approached the target regions of vowels.
>>
NEXT