README.md

---
license: cc-by-nc-4.0
tags:
- music
- documents
- end-to-end
- system-level
annotations_creators:
- manually expert-generated
pretty_name: JAZZMUS Staff Level
size_categories:
- 1K<n<10K
task_categories:
- image-to-text
- text-retrieval
subtasks:
- document-retrieval
extra_gated_fields:
  Affiliation: text
dataset_info:
  features:
  - name: image
    dtype: image
  - name: transcription
    dtype: string
  splits:
  - name: train
    num_bytes: 97519660.608
    num_examples: 1693
  - name: validation
    num_bytes: 5052012.0
    num_examples: 100
  - name: test
    num_bytes: 12733732.0
    num_examples: 228
  download_size: 111864542
  dataset_size: 115305404.608
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
---
🚨 Please check the original dataset page for further information. 
🤗 [Original JAZZMUS dataset card](https://huggingface.co/datasets/PRAIG/JAZZMUS)

# Optical Music Recognition of Jazz Lead Sheets - Staff Level Dataset
We provide musical scores for 163 unique jazz standards in MusicXML and Humdrum **kern format. The latter is widely used in systems that output musical scores, because it is a compact and easy-to-handle format.
The MusicXML scores are sourced from the Wikifonia database (discontinued in 2013) and have been partially corrected. We also leave the lyrics, if present in the original files, as they could be helpful for future extensions, but we don't consider them in this work.

## How to use it
Check the following example.

```python
from datasets import load_dataset


DATASET_NAME = "PRAIG/JAZZMUS_staffLevel"

ds = load_dataset(DATASET_NAME)

image = ds["train"][0]["image"]
transcription = ds["train"][0]["transcription"]
print(f"Image: {type(image)}")
print(f"Transcription: {repr(transcription)}")
```