README.md

---
task_categories:
- translation
- automatic-speech-recognition
language:
- gl
- en
size_categories:
- 1K<n<10K
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
  - split: test
    path: data/test-*
dataset_info:
  features:
  - name: id
    dtype: int32
  - name: audio
    dtype:
      audio:
        sampling_rate: 16000
  - name: text_gl
    dtype: string
  - name: text_en
    dtype: string
  splits:
  - name: train
    num_bytes: 1760117840.7617247
    num_examples: 2587
  - name: validation
    num_bytes: 351751435.47072464
    num_examples: 517
  - name: test
    num_bytes: 235408103.8585507
    num_examples: 346
  download_size: 2338731336
  dataset_size: 2347277380.091
---
# Dataset Details

**FLEURS-SpeechT-GL-EN** is Galician-to-English dataset for Speech Translation task.

This dataset has been compiled from Google's **[FLEURS data set](https://huggingface.co/datasets/google/fleurs)**.
It contains ~10h11m of galician audios along with its text transcriptions and the correspondant English translations.

# Preprocessing

This dataset is based on Google's FLEURS speech dataset, by aligning English and Galician data.
The alignment process has been performed following **[ymoslem's FLEURS dataset processing script](https://github.com/ymoslem/Speech/blob/main/FLEURS-GA-EN.ipynb)**

### English translations quality

To get a sense of the quality of the english text with respect to the galician transcriptions, a Quality Estimation model has been applied.

- **QE model**: [Unbabel/wmt23-cometkiwi-da-xl](https://huggingface.co/Unbabel/wmt23-cometkiwi-da-xl)
- **Average QE score**: 0.76

# Dataset Structure

```
DatasetDict({
    train: Dataset({
        features: ['id', 'audio', 'text_gl', 'text_en'],
        num_rows: 3450
    })
})
```

# Citation

```
@article{fleurs2022arxiv,
  title = {FLEURS: Few-shot Learning Evaluation of Universal Representations of Speech},
  author = {Conneau, Alexis and Ma, Min and Khanuja, Simran and Zhang, Yu and Axelrod, Vera and Dalmia, Siddharth and Riesa, Jason and Rivera, Clara and Bapna, Ankur},
  journal={arXiv preprint arXiv:2205.12446},
  url = {https://arxiv.org/abs/2205.12446},
  year = {2022},
```

Yasmin Moslem preprocessing script: https://github.com/ymoslem/Speech/blob/main/FLEURS-GA-EN.ipynb

## Dataset Card Contact

Juan Julián Cea Morán (jjceamoran@gmail.com)