ArEEG_Words

Description

This dataset is ArEEG_Words, an Arabic envisioned-speech (imagined-speech) EEG corpus introduced by a group at the Department of Information and Communication Engineering, Arab International University, Daraa, Syria (Darwish et al., 2024). It was built to address the scarcity of publicly available EEG datasets for non-English, low-resource languages: the authors describe it as the first Arabic word-level EEG dataset for brain—computer interface (BCI) research. 22 university-educated, Arabic-native adults wore an Emotiv EPOC X consumer headset and, on cue, silently imagined (“envisioned”) one of 16 commonly used Arabic words — a vocabulary of directional and assistive commands (up, down, left, right, yes, no, on, off, select, alarm, delete, space, hunger, thirst, medicine, bathroom). The downstream intent is to train deep-learning models that decode imagined Arabic words from scalp EEG for communication-impaired patients (Darwish et al., 2024).

Citation

• Paper DOI: 10.48550/arXiv.2411.18888
• Dataset DOI: 10.17632/7m472ykkx7.1
• Reference: Darwish, H., Al Malah, A., Al Jallad, K., & Ghneim, N. (2024). ArEEG_Words: Dataset for Envisioned Speech Recognition using EEG for Arabic Words. arXiv:2411.18888. Dataset: Mendeley Data V1.

Catalog citation block

Darwish, H., Al Malah, A., Al Jallad, K., & Ghneim, N. (2024). ArEEG_Words: Dataset for Envisioned Speech Recognition using EEG for Arabic Words. arXiv:2411.18888. DOI 10.48550/arXiv.2411.18888.

Dataset record: Darwish, H.; Almalah, A.; Al Jallad, K.; Ghneim, N. (2024). ArEEG_Words (V1) [Data set]. Mendeley Data. DOI 10.17632/7m472ykkx7.1. License: CC BY 4.0.

Download and access

Field	Value
Official source	https://data.mendeley.com/datasets/7m472ykkx7/1
Access class	open
Size	TODO(confirm)
Format	csv
BIDS	false
Mirrors	none recorded

Source & access

The dataset is archived on Mendeley Data: Darwish, H.; Almalah, A.; Al Jallad, K.; Ghneim, N. (2024), “ArEEG_Words”, Mendeley Data, V1, DOI 10.17632/7m472ykkx7.1 (landing page https://data.mendeley.com/datasets/7m472ykkx7/1), published 2 December 2024. The accompanying data-descriptor preprint is on arXiv: arXiv:2411.18888 (DOI 10.48550/arXiv.2411.18888, submitted 28 November 2024). The Mendeley record is released under the Creative Commons Attribution 4.0 (CC BY 4.0) license.

Data structure

The dataset is distributed as plain CSV (one file per recording, EmotivPRO export), not BIDS. Filenames follow the pattern par.<N> <arabic-word>_EPOCX_<serial>_<ISO-timestamp>.md.csv. Each file begins with a one-line EmotivPRO metadata header (title, start/stop timestamps, headset serial and firmware, channel count, sampling rates, sample count, version) followed by the column header and samples. No preprocessed/derivative tier is provided — the release is raw exports. No git-annex/DataLad packaging is used.

License

• Name: CC-BY-4.0
• Snippets on this site: allowed — the site may ship short derived snippets and precomputed products from this dataset, labelled as derivatives (re-referenced, filtered, cropped) with citation, license and DOI.

Acquisition

Field	Value
Device	14-channel wireless saline-electrode headset (Emotiv EPOC X)
Device class	consumer
Channels	14
Sampling rate (Hz)	128
Online filters	device factory front-end; not documented in the paper
Reference	CMS/DRL reference electrodes above the ears
Mains frequency (Hz)	TODO(confirm)
Paradigms	imagined-speech
Participants	22
Population	22 Arabic-speaking adults 17–25 (17 M / 5 F)
Clinical groups	none
Sessions	1
Durations	16 words × 10-s imagine with 20-s rest; ~1 h per participant; 352 recordings

Acquisition

Recording used the Emotiv EPOC X wireless headset with 14 EEG channels placed per the international 10—20 system: AF3, F7, F3, FC5, T7, P7, O1, O2, P8, T8, FC6, F4, F8, AF4, plus two reference electrodes (CMS and DRL, positioned above the ears) (Darwish et al., 2024). Signals were captured internally at 2048 Hz and down-sampled to 128 Hz per channel for export, streamed to the host computer over Bluetooth. Each CSV carries 67 columns: the 14 EEG channels (EEG.AF3 … EEG.AF4), per-channel contact-quality and signal-quality metrics (CQ.*, EQ.*), a 9-axis motion block (MOT.* — quaternions, accelerometer, magnetometer), markers, battery, and timestamps. Online filtering/reference settings beyond the EPOC X factory front-end are not documented in the paper.

Participants

The data-descriptor reports 22 participants (Table 2: Par.1—Par.22), 17 male / 5 female (~70% male), all educated to university level and Arabic-native speakers (Darwish et al., 2024). Reported mean age ≈ 22 years (the participant table averages 22.5; tabulated ages span 17—25 years). The abstract’s prose claim of an age range “between 16 and 64” appears to be a figure-axis error inconsistent with the participant table, so the table-based 17—25 range should be preferred. Participants were instructed to avoid caffeine, alcohol, and smoking for 8 hours before recording to limit nervous-system effects, and to keep calm with clear thoughts throughout (Darwish et al., 2024). The cohort is healthy (no clinical group).

Tasks / conditions

The single paradigm is imagined (envisioned) speech of Arabic words. Per participant, each of the 16 word slides was shown in randomized order; following the 10-second display of a word, the participant imagined the word with eyes closed for 10 seconds, then took a 20-second break to reset before the next word (Darwish et al., 2024). Each session lasted about one hour, and experimenters periodically confirmed the participant was maintaining focus. This protocol yielded a documented 352 EEG recordings (≈ 22 participants × 16 words). For machine-learning use the descriptor segments each recording into 250 ms windows, producing 15,360 EEG signals in total. The 16 classes (filenames are in Arabic): نعم (yes), لا (no), يمين (right), يسار (left), أعلى (up), أسفل (down), تشغيل (on/start), اطفاء (off), اختر (select), انذار (alarm), حذف (delete), مسافة (space), جوع (hunger), عطش (thirst), دواء (medicine), حمام (bathroom).

Used on this site

L7.5

Loader

Fetches only the stated subset. Recorded loader: TODO(confirm).

TODO(confirm) — the catalog records no loader for this dataset.

Caveats worth knowing

• Nominal 128 Hz output but ≈ 43 Hz hardware bandwidth and built-in 50/60 Hz notches — the Nyquist frequency is not the usable bandwidth.

Notable caveats

• Paradigm and naming. This is imagined/inner speech, not overt voice commands. There are several distinct “ArEEG” datasets — ArEEG_Chars (Arabic characters, EPOC X, arXiv:2402.15733), this ArEEG_Words, and the unrelated Scientific Data “ArEEG: an Open-Access Arabic Inner Speech EEG Dataset” (g.tec Unicorn, 8 channels, OpenNeuro ds005262) — do not conflate them.
• Subject count vs. device count. The paper documents 22 participants.
• Effective bandwidth. Although exported at 128 Hz (Nyquist 64 Hz), the EPOC X hardware front-end rolls off well below that (manufacturer effective bandwidth ≈ 0.16—43 Hz) with built-in mains notches.
• Consumer-grade montage. 14 fixed frontal/temporal/parietal/occipital saline-felt electrodes give coarse spatial coverage relative to research caps; contact-quality (CQ.*/EQ.*) columns should be screened before analysis.
• No formal demographic/clinical metadata file beyond the in-paper participant table; per-recording labels live only in the filename.

Related datasets

• MultiPENG (EEG stream)
• EEG for natural-image recognition (VEP)
• EEGEmotions-27
• PhySF, Physiological Sense of Flow
• ERP CORE (Compendium of Open Resources and Experiments)

# Loader: TODO(confirm) — the catalog records no loader for this dataset.
# Download a small subset from the official source: https://data.mendeley.com/datasets/7m472ykkx7/1