README.md

---
license: mit
task_categories:
- text-generation
pretty_name: IrishMAN
size_categories:
- 100K<n<1M
tags:
- music
---

## Dataset Summary

The **Irish Massive ABC Notation (IrishMAN)** dataset consists of 216,284 Irish music tunes represented in the widely adopted [ABC notation](https://abcnotation.com/) format. This format utilizes ASCII text to encompass a comprehensive set of musical symbols, including pitches, rhythms, and essential structural elements such as section boundaries.

Curated from thesession.org and abcnotation.com, prominent platforms for sharing folk and traditional music, the dataset includes 214,122 tunes for training and 2,162 tunes for evaluation. Notably, all tunes are automatically annotated with control codes, providing valuable representations of their musical forms and structures. Additionally, a [subset](https://huggingface.co/datasets/sander-wood/irishman/blob/main/harmony.json) of the tunes (over 40K) includes chord symbols, further enhancing the dataset's potential for various applications, including melody harmonization and harmonic analysis.

As a comprehensive resource, the IrishMAN dataset caters to researchers in musicology and artificial intelligence, offering insights into the exploration and understanding of Irish traditional music. Furthermore, all tunes in the dataset are non-copyrighted, ensuring its ethical and legal usage for research and creative projects. However, it is important to be aware that the automatic extraction of control codes for musical forms may introduce some limitations in accuracy.

## Control Codes

Inspired by [CTRL](https://huggingface.co/ctrl), we incorporate control codes into [TunesFormer](https://huggingface.co/sander-wood/tunesformer) to represent musical forms. These codes, positioned ahead of the ABC notation, enable users to specify the structures of the generated tunes. The following control codes are introduced:

- **B:number of bars**: This code enables users to define the desired number of bars within a section. It utilizes the bar symbol `|` to establish the length of each section. The dataset's section length ranges from 1 to 32 bars (e.g., `B:1` for a one-bar section, and `B:32` for a section with 32 bars).

- **S:number of sections**: With this control code, users can determine the overall structure of the generated melody by specifying the number of sections it should contain. Various symbols represent section boundaries, including `[|`, `||`, `|]`, `|:`, `::`, and `:|`. The dataset's range for the number of sections is from 1 to 8 (e.g., `S:1` for a single-section melody, and `S:8` for a melody with eight sections).

- **E:edit distance similarity**: This control code provides fine-grained control over the similarity between the current section ($c$) and a previous section ($p$) in the generated melody. Similarity is calculated using the Levenshtein distance, measuring the difference between two sequences. The edit distance similarity ($eds(c,p)$) is calculated as:
  ```
  eds(c,p) = 1 - lev(c,p) / max(|c|,|p|)
  ```
  where $|c|$ and $|p|$ are the string lengths of the two sections. The control code has 11 levels, ranging from no match to an exact match (e.g., `E:0` for no similarity, and `E:10` for an exact match).

## Copyright Disclaimer
The dataset is provided solely for research purposes and is not intended for commercial use. 

If you are the copyright owner of any tune included in the IrishMAN and have concerns about its inclusion, please contact us at [email protected] to have it removed.

## Special Thanks
We would like to extend a special thanks to abcnotation.com and thesession.org for their contributions to the development and promotion of ABC notation, as well as their contributions to the field of music information retrieval. Their platforms have provided invaluable resources for the traditional and folk music community, and have made it possible for researchers like us to create and study large datasets like IrishMAN.