This study presents the results of a pilot study of the potential use of a portable low-channel EEG system for recording and discriminating brain states. The work consists of two stages: 1) comparing different EEG recording configurations (with and without conductive gel), 2) developing and testing a speech rehabilitation paradigm using the optimal configuration (based on the first stage). We demonstrated that recordings without gel significantly impede EEG pattern classification accuracy. Then, using the word-picture matching paradigm developed in our laboratory, we examined evoked responses and the feasibility of distinguishing them using machine learning methods. We found that the word-picture matching paradigm elicits a distinct evoked potential in the EEG of healthy participants. Furthermore, to minimize the classifier calibration time for each individual participant, we attempted to pretrain the classifier using data from independent participants. However, pretraining the classifier did not lead to a significant improvement in classification performance. Although further improvement is required to increase classification performance, this pilot study is one of the few aimed at the practical application of these results in rehabilitation practice.