EEG-based brain-computer interface (BCI) dataset for directional word recognition

This dataset contains multichannel EEG recordings collected during overt and covert (inner) articulation of six spatial-direction words. Data were acquired from 22 neurologically healthy volunteers: 12 native Russian speakers and 10 native Spanish speakers, aged 22–43 years. Recording setup: Monopolar EEG signals were recorded from 38 electrodes positioned according to the international 10–10 system, alongside simultaneous EMG recordings from the masseter muscle and laryngeal region. Data were acquired using a Neurovisor-BMM-52 (NVX) EEG system at a sampling rate of 500 Hz. Experimental task: Participants performed overt and covert articulation of six directional words: "up," "down," "left," "right," "forward," and "backward" (Russian: вверх, вниз, влево, вправо, вперёд, назад; Spanish: arriba, abajo, derecha, izquierda, adelante, atrás). Each participant produced approximately 110 ± 20 trials per word across both conditions. Dataset contents: Raw continuous EEG/EMG recordings in .edf format (38 EEG) Preprocessed epoched data in .fif format (MNE Epochs objects, dimensions: trials × 38 channels × 750 time points) Event files in .xlsx format with trial timestamps and condition labels Participant metadata in subject_metadata.json (age, native language) ANOVA results for spectral power analysis across conditions Structure:Inner Speech Dataset/├── Spanish/│ ├── sub0/│ │ ├── sub0-epo.fif│ │ ├── sub0.edf│ │ └── sub0.xlsx│ ├── ...│ └── sub9/│ ├── sub9-epo.fif│ ├── sub9.edf│ └── sub9.xlsx├── Russian/│ ├── sub1/│ │ ├── sub1-epo.fif│ │ ├── sub1.edf│ │ └── sub1.xlsx│ ├── ...│ └── sub12/│ ├── sub12-epo.fif│ ├── sub12.edf│ └── sub12.xlsx└── subject_metadata.json Preprocessing: Bandpass filtering (1–70 Hz), 50 Hz notch filter, ICA-based artifact removal. Ethics: The study was approved by the Committee of Bioethics of Southern Federal University (Report No. 3, Application No. 7, dated September 9, 2022). A consent waiver for open data sharing was granted, confirming that individual participant consent is not required as the dataset contains no direct identifiers. Potential applications: Development and benchmarking of BCI algorithms based on inner speech, cross-linguistic EEG analysis, neurolinguistic research, real-time neurocommunication systems.