# JapanEEG Open-Vocabulary Japanese Sentence Reading EEG-EMG Dataset from OpenNeuro ([ds007602](https://openneuro.org/datasets/ds007602)). - **3 long-term participants** with 600+ hours of EEG recorded during overt Japanese speech production - **Open vocabulary** -- participants read sentences from novels, text-based TV games, and the JSUT corpus at natural speed, producing long-form speech - **Multiple devices** -- g.PANGOLIN (128 ch), g.SCARABEO (64 ch), eego™ sports (61 ch), each with EOG + 2x EMG channels ## Quick Start ### 1. Setup ```bash git clone https://github.com/Motoshige496/JapanEEG.git cd JapanEEG uv sync ``` ### 2. Download the Dataset ```bash # Download all data uv run python -m japaneeg.download # Or download a single subject/session for quick testing uv run python -c " from japaneeg.download import download download(include=['sub-01/ses-20230829/eeg/*run-01*']) " ``` ### 3. Load and Explore EEG Data ```python import mne import mne_bids import pandas as pd # --- Load via MNE-BIDS --- bids_path = mne_bids.BIDSPath( subject="01", session="20230829", task="speechopen", acquisition="pangolin", run="01", suffix="eeg", datatype="eeg", root="data/ds007600", ) raw = mne_bids.read_raw_bids(bids_path, verbose=False) print(raw.info) # # --- Read events --- events_tsv = bids_path.copy().update(suffix="events", extension=".tsv") events = pd.read_csv(str(events_tsv), sep="\t") print(events.head()) # onset duration trial_type sample value # 0 20.0 5.0 speech_window 20480 ルーシーは伝統の # 1 25.0 5.0 speech_window 25600 スイッチをつけた # --- Plot raw EEG (first 10 seconds) --- raw.load_data() raw.pick("eeg") raw.plot(duration=10, n_channels=20, title="JapanEEG - sub-01") ``` ## Dataset Structure ``` data/ds007600/ ├── dataset_description.json ├── participants.tsv ├── README ├── sub-{01,02,03,04}/ │ └── ses-YYYYMMDD/ │ └── eeg/ │ ├── *_eeg.edf # EEG recording (EDF format) │ ├── *_eeg.json # EEG metadata │ ├── *_channels.tsv # Channel information │ └── *_events.tsv # Speech events with Japanese text ``` ## License - **Code**: CC0-1.0 - **Dataset**: CC0-1.0