Silent speech BCI

点击翻译按钮获取中文摘要

A new silent speech BCI was proposed using both EEGs scalp-recorded and speech signals measured during overtly and covertly speaking "janken" and "season" in Japanese. The framework consists of two phases. The first phase specifies a Kalman filter (KF) using both spectrograms of the speech signals and independent components of the EEGs recorded during the actual speech task. In case of the "season" task, speech signals were transformed into vowel and/or consonant sequences, and these relations were learned by HMM. The second one predicts spectrograms for the silent "janken" and "season" by the KF with the EEGs during the silent speech tasks. For the silent "season", the predicted spectrograms were inputted to the HMM. The present results demonstrate that when using the Kalman filter with the ICs whose dipole solutions are localized to mainly the Broca's area, the silent "jankens" are all correctly predicted for ten healthy subjects, and the "spring"- and "summer"-HMMs worked well.