Imagined Speech Decoding From EEG: The Winner of 3rd Iranian BCI Competition (iBCIC2020)

Abstract

Brain-computer interface (BCI) is defined as the combination of machine and brain signals to control a device or computer to improve the quality of life, e.g., for people with paralysis. In this paper, we focus on people with speech disorders and investigate the capability of electroencephalogram (EEG) signals to discriminate four classes, including the speech imagination of three Persian words corresponding to the English words "rock," "paper," and "scissors," in addition to the resting state. We used the data available from the 3rd Iranian BCI competition (iBCIC2020), acquired from 10 healthy participants in a randomized study. Initially, the mutual information (MI) was used to find the optimum frequency band. Then, features were extracted from the data using the Common Spatial Pattern (CSP) algorithm. Afterward, the most discriminative features were selected using the neighborhood component analysis (NCA). These features were then fed to a meta-classifier based on the stacking ensemble learning. The results show that working on an optimum frequency band will enhance the results compared with the fixed frequency band. It is also worth mentioning that the optimum frequency band is subject dependent; therefore, it is substantial to be selected accurately. Our method achieved an average classification accuracy of 51.90%±2.73 across all participants, which is promising compared with the results of previous studies in the field of imagined speech recognition in subject dependent BCI systems with randomized order of the stimuli.