---
license: cc-by-nc-4.0
configs:
  - config_name: default
    data_files:
      - split: train
        path: "metadata.csv"
task_categories:
  - image-to-3d
  - other
language:
  - cy
tags:
  - 3d-face
  - 3d-mesh
  - facial-reconstruction
  - 4d-dataset
  - texture-maps
  - welsh
  - cymraeg
  - multimodal
size_categories:
  - 100K<n<1M
pretty_name: CymruFluency Welsh Speech Dataset - 3D Meshes
---

# Welsh Speech Dataset - 3D Facial Meshes

3D facial reconstructions from the Welsh Speech Dataset.

## Contents

- **3D meshes** (.obj files) - One per frame
- **Texture maps** (.png files) - Fused left-right stereo images from 3DMD
- Captured using 3DMD 6-camera system
- **~330 zip files** (one per speaker-phrase sequence)

## File Structure

Files are organized as zip archives in the `meshes/` directory, one zip per speaker-phrase sequence:

```
meshes/
├── speaker_01_phrase_01.zip
├── speaker_01_phrase_02.zip
├── ...
└── speaker_33_phrase_10.zip
```

Each zip contains frame files with simple naming:

```
speaker_01_phrase_01.zip
├── 001.obj
├── 001.png
├── 002.obj
├── 002.png
└── ...
```

## Texture Maps

The PNG files contain fused left-right stereo image pairs captured by the 3DMD system, side-by-side in a single image.

## Preview

<table>
  <tr>
    <td><img src="example-3d-reconstructed.gif" width="400"></td>
    <td><img src="example-texture-map.bmp" width="400"></td>
  </tr>
  <tr>
    <td align="center">3D Mesh Rotation</td>
    <td align="center">Texture Map</td>
  </tr>
</table>

## Metadata

See `metadata.csv` for sequence-level information:

| Column | Description |
|--------|-------------|
| `speaker_id` | Speaker identifier (1-33) |
| `phrase_id` | Phrase identifier (1-10) |
| `zip_path` | Path to zip file |
| `num_frames` | Number of frames in the zip |
| `fluency_score` | Pronunciation quality score (0-5) |
| `welsh_text` | Welsh phrase text |
| `english_translation` | English translation |

## Related Repositories

- **Main**: [arvinsingh/welsh-speech-dataset](https://huggingface.co/datasets/arvinsingh/welsh-speech-dataset)
- **Audio**: [arvinsingh/welsh-speech-audio](https://huggingface.co/datasets/arvinsingh/welsh-speech-audio)
- **Landmarks**: [arvinsingh/welsh-speech-landmarks](https://huggingface.co/datasets/arvinsingh/welsh-speech-landmarks)

## Usage

```python
from huggingface_hub import hf_hub_download
import zipfile
import os

# download specific sequence zip
zip_path = hf_hub_download(
    repo_id="arvinsingh/welsh-speech-3d-meshes",
    filename="meshes/speaker_01_phrase_01.zip",
    repo_type="dataset"
)

# extract
extract_dir = "speaker_01_phrase_01"
with zipfile.ZipFile(zip_path, 'r') as zf:
    zf.extractall(extract_dir)

for f in sorted(os.listdir(extract_dir)):
    print(f)  # 001.obj, 001.png, 002.obj, ...
```

### Working with meshes in memory (without extracting)

```python
import zipfile
import trimesh  # pip install trimesh
from io import BytesIO

with zipfile.ZipFile(zip_path, 'r') as zf:
    # read specific frame's mesh
    obj_data = zf.read("001.obj")
    mesh = trimesh.load(BytesIO(obj_data), file_type='obj')
    print(f"Vertices: {len(mesh.vertices)}, Faces: {len(mesh.faces)}")
```

## Citation

If you use this dataset, please cite both the paper and the dataset:

```bibtex
@inproceedings{bali_2026_cymrufluency,
  author       = {Bali, Arvinder Pal Singh and Tam, Gary KL and Siris, Avishek and Andrews, Gareth and Lai, Yukun and Tiddeman, Bernie and Ffrancon, Gwenno},
  title        = {CymruFluency - A Fusion Technique and a 4D Welsh Dataset for Welsh Fluency Analysis},
  booktitle    = {Advanced Concepts for Intelligent Vision Systems},
  pages        = {96--108},
  year         = 2026,
  publisher    = {Springer Nature Switzerland},
  doi          = {10.1007/978-3-032-07343-3_8},
}

@dataset{bali_2025_dataset,
  author       = {Bali, Arvinder Pal Singh and Tam, Gary KL and Siris, Avishek and Andrews, Gareth and Lai, Yukun and Tiddeman, Bernie and Ffrancon, Gwenno},
  title        = {Dataset and code for "CymruFluency"},
  year         = 2025,
  publisher    = {Zenodo},
  doi          = {10.5281/zenodo.15397513},
}
```