---
license: mit
---

# 🐶 Barkopedia Challenge Dataset  
[🔗 Barkopedia Website](https://uta-acl2.github.io/Barkopedia/)

## 📦 Dataset Description

This challenge provides a labeled dataset of dog bark audio clips for understanding the **arousal** and **valence** of emotional state from sound.

### 📁 Current Release

- **Training Set**
  - Includes:
    - `train/husky` and `train/shiba` contain all training audio clips for each of the two breeds
    - `husky_train_labels.csv` and `shiba_train_labels.csv` contain labels for all training audio clips
  - Total: **1000** training audio clips (500 for each breed)

- **Test Set**
  - To be released in **June**
  - Will contain **200** audio clips
    - **50% public (100 clips)** — used for live leaderboard updates
    - **50% private (100 clips)** — used for final evaluation

## 🔖 Labels

Each clip is annotated with **one arousal** and **one valence** category.

### 🎯 Arousal Categories (`arousal`)
- `Low`  
- `Medium`  
- `High`  


### 🌍 Valence Categories (`valence`)
- `Negative`  
- `Neutral`  
- `Positive`  



## 🚀 Submission & Leaderboard

All competition data, result submission, and the live leaderboard are hosted on **[Hugging Face](https://huggingface.co/)**.  
If you don’t have a Hugging Face account yet, please [register here](https://huggingface.co/join).

---

📌 *Stay tuned for the test set release in June and leaderboard launch!*

---