Single-trial analysis of voice-stimulus-evoked potentials

We are developing a brain-computer interface (BCI) for event-related potentials (P300) using speech stimulus in the Japanese language based on the need to investigate amyotrophic lateral sclerosis (ALS) patients. Previously, we studied a single-trial analysis of P300 with a 4-Hz lowpass filter in order to improve the entry speed of the BCI. However, the problem was a low detection accuracy, i.e., approximately 30%–80%. In this article, we reviewed the application of independent component analysis (ICA) in order to improve the accuracy of single-trial analysis of P300. As a result, the detection ratio improved from 54.2% for the traditional 4-Hz low-pass filter to 90.9% in the choice of one between two. Furthermore, in an off-line experiment, the detection ratio of the P300 response to each sound of “a, i, u, e and o” improved in the task to choose one among five with synthetic speech stimulus. The maximum detection ratio was 94.7%, and the detection ratio per sound improved from 47.0% to 85.1%.