omarfirdaus101/emg-cyton-digits

EMG Cyton Digit Recordings

Raw EMG recordings from two subjects performing spoken/written digit gestures (0–9), captured with an OpenBCI Cyton board via BrainFlow at 250 Hz on 8 channels.

Dataset summary

Field	Value
Subjects	2 Males, 18-25
Included recordings	18
Excluded partial recordings	7
Classes	10 digits (0–9)
Sampling rate	250 Hz
Channels	8 (emg_ch1 … emg_ch8, Cyton EEG pins used as EMG)
Session date	2026-06-29

Labeling

Each recording is labeled from the innermost digit folder in the original source path (e.g. .../Five/Recordings/... → digit 5 / "five").

Recordings shorter than 60000 samples (240 seconds) were excluded as incomplete sessions.

Excluded source files

See manifest.json for the full list of 7 partial recordings that were excluded (Omar digits 0, 4, 8 and short retakes).

Load with Hugging Face Datasets

from datasets import load_from_disk

dataset = load_from_disk("data")
print(dataset)
print(dataset[0]["digit"], dataset[0]["digit_name"], dataset[0]["num_samples"])

# Stack channels into (8, T)
import numpy as np
sample = dataset[0]
emg = np.stack([sample[f"emg_ch{i}"] for i in range(1, 9)])

After cloning from the Hub:

from datasets import load_dataset

dataset = load_dataset("your-username/emg-cyton-digits")

Upload to the Hub

From the parent of this folder:

pip install huggingface_hub datasets
huggingface-cli login
huggingface-cli upload your-username/emg-cyton-digits dataset/ --repo-type dataset

Or push programmatically:

from datasets import load_from_disk

ds = load_from_disk("dataset/data")
ds.push_to_hub("your-username/emg-cyton-digits")

Then upload the dataset card separately if needed:

huggingface-cli upload your-username/emg-cyton-digits dataset/README.md --repo-type dataset
huggingface-cli upload your-username/emg-cyton-digits dataset/manifest.json --repo-type dataset

Raw source format

Original files were BrainFlow tab-separated CSV exports with 24 columns per time sample (8 EEG/EMG channels, accelerometer, timestamps, marker channel). See BrainFlow Cyton docs for board layout.

Suggested evaluation

With only two subjects, prefer leave-one-subject-out cross-validation rather than random splits.

Citation

If you use this dataset, please cite Omar Firdaus of Mausix Research.