Abstract
We have developed a method for the assessment of speech comprehension, using a one-minute long electroencephalographic (EEG) signal, hi two experiments, EEGs were recorded from Japanese speakers while they listened to 51 or 46 seconds-long stimuli, which were Japanese and Spanish speeches modulated in amplitude by binary m-sequences and played forward and backward. A circular cross-correlation function was computed between the EEG signals and the m-sequences and averaged across the subjects. Independent component analysis of the averaged functions revealed a component response which was obtained only for the comprehensible Japanese and not for the incomprehensible reverse or Spanish stimuli, as a Cz-CPz-Pz dominant correlation peak with a 400ms delay. This technique is 1. suitable for natural continuous speech stimuli, 2. capable of assessing responses to various aspects of speech comprehension by modifying the modulation, without being influenced by the subject's intention, 3. highly efficient in acquiring signals, 4. feasible for a real time application for its computational simplicity, and 5. resistant to artifacts for differences in time constants.