Open datasets

• IntracorticalOpen

  Brain-to-Text '25

  Intracortical recordings of attempted speech from the participant in the 2024 NEJM speech study, released as a public decoding benchmark.

  Task

  Attempted speech of sentences

  Participants

  1 (ALS)

  Format

  HDF5

  Host

  Dryad; Kaggle competition

• IntracorticalOpen

  Brain-to-Text Benchmark '24

  The dataset behind the first large-vocabulary intracortical speech BCI, with a held-out test set scored by an evaluation server.

  Task

  Attempted speech of sentences

  Participants

  1 (ALS)

  Format

  MATLAB files

  Host

  Dryad

• IntracorticalOpen

  Handwriting BCI dataset

  Neural activity recorded while a participant attempted to handwrite letters and sentences.

  Task

  Attempted handwriting

  Participants

  1 (spinal cord injury)

  Format

  MATLAB files

  Host

  Dryad

• IntracorticalOpen

  FALCON benchmark

  A set of human, monkey and songbird datasets for testing whether decoders stay accurate across days with little recalibration.

  Task

  Movement, handwriting and vocalisation across sessions

  Participants

  Human and non-human datasets

  Format

  NWB

  Host

  DANDI Archive

• MEG / OPMOpen

  LibriBrain / LibriBrain100

  Deep within-subject MEG recorded while listening to audiobooks, extended in 2026 with 32 additional subjects for cross-subject decoding.

  Task

  Listening to naturalistic speech

  Participants

  1 subject with about 80 hours; 32 further subjects with about 40 minutes each

  Format

  Standardised MEG files with event annotations

  Host

  PNPL, University of Oxford

• ECoGOpen

  Podcast ECoG

  Electrocorticography from participants listening to a 30-minute podcast, organised in BIDS for language-model analyses.

  Task

  Natural language comprehension

  Participants

  9 (epilepsy monitoring)

  Format

  BIDS

  Host

  OpenNeuro

• ECoGOpen

  AJILE12

  Long-term ECoG with pose tracking during naturalistic, unstructured behaviour in hospital.

  Task

  Naturalistic arm movements

  Participants

  12 (epilepsy monitoring)

  Format

  NWB

  Host

  DANDI Archive

• EEGOpen

  BNCI 2014-001 (BCI Competition IV 2a)

  The most widely used motor-imagery benchmark: four imagined movements recorded with 22 EEG channels.

  Task

  Four-class motor imagery

  Participants

  9

  Format

  GDF / MAT

  Host

  BNCI Horizon 2020; accessible through MOABB

• EEGOpen

  EEG Motor Movement/Imagery

  A large-cohort EEG dataset of executed and imagined movements, recorded with the BCI2000 system.

  Task

  Motor execution and imagery

  Participants

  109

  Format

  EDF+

  Host

  PhysioNet

• sEMGOpen

  emg2qwerty

  Surface EMG from both wrists during touch typing, released with baselines for decoding keystrokes.

  Task

  Touch typing

  Participants

  108

  Format

  HDF5

  Host

  GitHub (facebookresearch)

Fields a source does not state are shown as “not stated” rather than estimated. Always read a dataset’s own licence and terms before use. See also the data standards these datasets are stored in.