pianistprogrammer/abc2vec-irish-folk

ABC2Vec Irish Folk Music Dataset

This dataset contains 211,524 Irish traditional tunes in ABC notation, preprocessed and split for training representation learning models.

Dataset Description

• —Curated by: IrishMAN Dataset (The Session + ABCnotation.com)
• —Processed for: ABC2Vec: Self-Supervised Representation Learning for Irish Folk Music
• —Language: ABC notation (symbolic music format)
• —License: CC-BY-4.0

Dataset Structure

Data Splits

Data Fields

Each tune contains:

• —tune_id: Unique identifier
• —title: Tune name
• —abc_body: ABC notation of the melody
• —tune_type: Rhythmic category (jig, reel, polka, waltz, etc.)
• —mode: Tonal mode (major, minor, dorian, mixolydian)
• —key: Key signature
• —meter: Time signature
• —bar_count: Number of bars in the tune

Dataset Statistics

• —Tune Types: 44.9% reels, 21.3% jigs, 14.5% polkas, 12.2% waltzes
• —Modes: 80.2% major, 11.3% minor, 5.4% Dorian, 3.0% Mixolydian
• —Keys: 30.5% G, 26.8% D, 13.9% A (sharp keys dominant)
• —Median Length: 18 bars, 287 characters

Usage

from datasets import load_dataset

# Load the entire dataset
dataset = load_dataset("pianistprogrammer/abc2vec-irish-folk-dataset")

# Access splits
train = dataset["train"]
val = dataset["validation"]
test = dataset["test"]

# Example tune
print(train[0]["abc_body"])
print(f"Type: {train[0]['tune_type']}, Mode: {train[0]['mode']}")

Citation

If you use this dataset, please cite:

@article{abc2vec2025,
  title={ABC2Vec: Self-Supervised Representation Learning for Irish Folk Music},
  author={[Your Name]},
  journal={[Journal Name]},
  year={2025}
}

Source

This dataset is derived from:

• —The Session (thesession.org): Community-maintained Irish traditional music archive
• —ABCnotation.com: Long-standing ABC notation repository

Processed as part of the IrishMAN (Irish Music ABC Notation) corpus.

License

Creative Commons Attribution 4.0 International (CC-BY-4.0)

The original tunes are traditional folk music in the public domain. This processed dataset is released under CC-BY-4.0.

Additional Files

• —vocab.json: Character vocabulary for tokenization (98 tokens)
• —metadata.csv: Complete metadata for all 211,524 tunes

Contact

For questions or issues with this dataset, please open an issue on the ABC2Vec GitHub repository.