Nguyen2017 — Imagined speech EEG for Riemannian manifold classification (re-hosted)

Re-hosted EEG data from Nguyen, Karavas & Artemiadis (2017), originally distributed through a Dropbox archive (dataset.zip) as a companion to the Journal of Neural Engineering paper. Repackaged into four normalized condition ZIPs for programmatic access from the MOABB benchmarking framework.

Paper

C. H. Nguyen, G. K. Karavas, and P. Artemiadis, “Inferring imagined speech using EEG signals: a new approach using Riemannian manifold features,” Journal of Neural Engineering, vol. 15, no. 1, 2017.
DOI: 10.1088/1741-2552/aa8235

Paradigm

Imagined speech in four experimental conditions:

Each trial uses a 5-second imagined-speech window (the last_beep interval in the authors' terminology).

Participants

• Subjects counts vary per condition (6–8); see table above and the per-condition README inside each zip. Some participants contributed to multiple conditions on different recording days.
• Institution: Arizona State University
• IRB: ASU Protocols 1309009601, STUDY00001345

Recording setup

File structure

Four condition ZIPs at the top level of this record. Each contains normalized per-subject .mat files and an embedded README with the original → clean filename mapping:

Vowels.zip
  README.md
  sub-01.mat   ... sub-08.mat
Short_words.zip
  README.md
  sub-01.mat   ... sub-06.mat
Long_words.zip
  README.md
  sub-01.mat   ... sub-06.mat
Short_Long_words.zip
  README.md
  sub-01.mat   ... sub-06.mat
README.md           ← top-level record readme

The .mat payloads are bit-identical to the authors' originals; only filenames inside the zips were normalized (from e.g. sub_4b_ch80_v_eog_removed_256Hz.mat to sub-01.mat). See each inner README for the full filename mapping.

Data format

Each .mat file contains a MATLAB variable eeg_data_wrt_task_rep_no_eog_256Hz_last_beep, a (n_classes, n_trials) object array whose cells are (n_channels, 1280) float matrices (5 s × 256 Hz).

Loading with MOABB

from moabb.datasets import (
    Nguyen2017_V, Nguyen2017_S, Nguyen2017_L, Nguyen2017_SL,
)
from moabb.paradigms import MotorImagery

ds = Nguyen2017_V()
paradigm = MotorImagery(
    events=["vowel_a", "vowel_i", "vowel_u"],
    n_classes=3,
)
X, y, metadata = paradigm.get_data(dataset=ds, subjects=[1])

Re-hosting rationale

The original Dropbox share relies on an rlkey token that can rate-limit and has no persistent DOI. Additionally, the outer dataset.zip wrapper has a 4 GB prefix that breaks Python's built-in zipfile module (it requires the system unzip tool to extract). This Zenodo mirror provides clean, DOI-addressed URLs for the four condition zips so MOABB's automated benchmarking pipeline can fetch the data without a workaround.

The signal data is unchanged — only the outer wrapper archive has been removed and filenames inside each condition zip were normalized.

License and attribution

Each of the authors’ original Read_me.txt files (included verbatim in this mirror) states: “This data is provided by the Human-Oriented Robotics and Controls (HORC) lab, ASU. When using this dataset, please cite the following paper: C. Nguyen, G. Karavas, P. Artemiadis, ‘Inferring imagined speech using EEG signals…’, Journal of Neural Engineering, July 2017.”

The authors grant no explicit Creative Commons license, only a citation request. This mirror is published under Zenodo’s other-open category on the same implicit terms as the authors’ Dropbox distribution: open for research use with attribution to the paper (10.1088/1741-2552/aa8235). If you are one of the dataset authors and wish to clarify the license or request removal of this mirror, please contact chuong.h.nguyen@asu.edu, panagiotis.artemiadis@asu.edu, or file an issue at NeuroTechX/moabb.