Pressel2016 — Imagined speech EEG (Spanish vowels and directional commands, re-hosted)

Re-hosted EEG data from Pressel Coretto, Gareis, and Rufiner (2017), originally distributed through Google Drive (file id 0By7apHbIp8ENZVBLRFVlSFhzbHc) as an open-access companion to the SIPAIM/SPIE proceedings paper. Repackaged into per-subject ZIP archives for easier programmatic access from the MOABB benchmarking framework.

Paper

G. A. Pressel Coretto, I. E. Gareis, and H. L. Rufiner, “Open access database of EEG signals recorded during imagined speech,” in Proc. SPIE 10160, 12th International Symposium on Medical Information Processing and Analysis, 2017.
DOI: 10.1117/12.2255697

Paradigm

11-class imagined speech of Spanish phonemes and directional commands:

• Vowels: /a/, /e/, /i/, /o/, /u/ (stim codes 1-5)
• Directional commands: arriba (up), abajo (down), adelante (forward), atrás (back), derecha (right), izquierda (left) (stim codes 6-11)

Two modalities were recorded per stimulus: imagined speech (modality code 1) and pronounced speech (modality code 2). The MOABB adapter loads the imagined condition by default.

Each trial is 4 seconds at 1024 Hz (4096 samples).

Participants

• 15 Argentinian volunteers (7 female, 8 male, ages 24-28)
• Institution: Universidad Nacional de Entre Ríos (Argentina)

Recording setup

File structure

15 per-subject ZIP files (S01.zip–S15.zip). Each contains a single MATLAB file renamed from the original S{NN}_EEG.mat to sub-{NN}_eeg.mat:

S01.zip
  sub-01_eeg.mat
S02.zip
  sub-02_eeg.mat
...
S15.zip
  sub-15_eeg.mat
README.md   ← top-level, stimulus codes and channels

The .mat files are bit-identical to the originals; only the filename was changed to adopt the MOABB subject-ID convention.

Data format

Each .mat file contains an EEG variable of shape (n_trials, 6 * 4096 + 3). The first 24576 columns are flattened channel-first EEG samples; the last three columns are labels: [modality, stimulus_code, artifact_flag].

Loading with MOABB

from moabb.datasets import Pressel2016
from moabb.paradigms import MotorImagery

dataset = Pressel2016()
paradigm = MotorImagery(
    events=["vowel_a", "vowel_e", "vowel_i"],
)
X, y, metadata = paradigm.get_data(dataset=dataset, subjects=[1])

Re-hosting rationale

The original Google Drive share relies on the gdown library and a legacy resourcekey token. Both are fragile: Google Drive can quota-block downloads, change its confirm-token protocol, or revoke resource keys (we observed all three modes of failure). This Zenodo mirror provides direct, DOI-addressed URLs so the data can be fetched from any CI environment without credentials and without depending on consumer cloud storage.

The signal data is unchanged — only filename conventions and archive layout have been normalized.

License and attribution

The original SPIE publication describes this dataset as “open access” and “publicly available” but does not declare an explicit Creative Commons license. This mirror is published under Zenodo’s other-open category on the same terms as the authors’ original distribution: open for research use with attribution to the paper (10.1117/12.2255697). If you are one of the dataset authors and wish to clarify the license or request removal of this mirror, please contact germanpressel@gmail.com (the corresponding author listed in the paper) or file an issue at NeuroTechX/moabb.