README.md

---
license:
- other
pretty_name: MIAO dataset
tags:
- audio
- image
- multimodal
- onomatopoeia
---

# 🐾 MIAO: Multimodal Image-Audio Onomatopoeia Dataset

## Dataset Summary

**MIAO** is a multimodal dataset consisting of paired **sound event clips** and **onomatopoeic images**, which is designed to support research and development on multimodal correspondence between sounds and visual onomatopoeic expressions.
It can be used for a wide range of tasks, including cross-modal retrieval, multimodal representation learning, and generative modeling of sounds and images.

### Examples of onomatopoeic images

<div style="display: flex; gap: 24px; justify-content: center; align-items: flex-start;">
  <div style="text-align: center;">
    <img src="sample/Chainsaw.png" alt="Chainsaw" width="300">
    <p style="margin-top: 0px;">Onomatopoeic image for "Chainsaw"</p>
  </div>
  <div style="text-align: center;">
    <img src="sample/Dog_barking.png" alt="Dog barking" width="300">
    <p style="margin-top: 0px;">Onomatopoeic image for "Dog barking"</p>
  </div>
</div>

## Citation

If you use this dataset, please cite the following paper:

Keisuke Imoto, Yamato Kojima, and Takao Tsuchiya, "Audio-Image Cross-Modal Retrieval with Onomatopoeic Images," arXiv:2605.17509, 2026.

```bibtex
@article{imoto2026audio,
  title={Audio-Image Cross-Modal Retrieval with Onomatopoeic Images},
  author={Imoto, Keisuke and Kojima, Yamato and Tsuchiya, Takao},
  journal={arXiv preprint arXiv:2605.17509},
  year={2026}
}
```

## Dataset Overview

| Data type | Source | \# sound event classes | \# per class | Total |
|---|---|---|---|---|
| Audio | FSD50K (CC0 clips) | 50 | 12 clips | 600 clips |
| Image | Original images | 50 | 18 images | 900 images |

The audio clips are organized into 12 audio sets. Each audio set contains one clip for each sound event class.
The images were created by 18 illustrators. Each illustrator created one onomatopoeic image for each sound event class after listening to the corresponding audio clip.

## Data Sources and Creation

### Audio

The audio clips were selected from the **CC0-licensed subset of FSD50K**.

### Images

The onomatopoeic images were created by 18 illustrators recruited through the 99designs platform.
Each illustrator listened to 50 sound event clips and created one onomatopoeic image for each sound event class.

## License

MIAO contains audio files and image files with different licensing terms.

### Audio files

The audio files are drawn from the CC0-licensed subset of FSD50K and are distributed under the same **CC0-1.0 license**.

### Image files

The onomatopoeic image files are available under the following terms:

- **Research use:** The image files may be used under the **MIT License** for research purposes.
- **Commercial use:** Commercial use of the image files requires a separate paid license.

For commercial use of the image files, please contact:

```text
Keisuke Imoto
keisuke.imoto (at_mark) ieee.org