---
license: cc-by-nc-4.0
task_categories:
- text-to-speech
pretty_name: TunArTTS
---

# Dataset Description: 

This speech corpus is extracted from an online English-Tunisian Arabic dictionary Derja Ninja, providing a valuable resource for linguistic and speech-related research. 
The dataset contains over 3 hours of mono-speaker audio recordings from a male speaker, sampled at 44.1 kHz.

Key characteristics of the corpus include:

- **Language**: Tunisian Arabic.
   
- **Speaker**: Single male speaker.
   
- **Sampling Rate**: High-quality recordings at 44.1 kHz.
   
- **Manual Diacritization**: All text has been processed and manually diacritized, ensuring phonetic accuracy for Tunisian Arabic.
 
This corpus is well-suited for applications such as speech synthesis and automatic speech recognition.

# Dataset Characteristics

| **Characteristic**         | **Value**                |
|-----------------------------|--------------------------------|
| Total Segments             | 1493                           |
| Total Words                | 20925                          |
| Total Characters           | 113221                         |
| Total Duration             | 3 hours and 32 seconds         |
| Mean Clip Duration         | 7.24 seconds                   |
| Min Clip Duration          | 3.11 seconds                   |
| Max Clip Duration          | 16.3 seconds                   |
| Mean Words per Clip        | 14.015                         |
| Distinct Words             | 4491                           |


A research paper based on this dataset has been published. You can find the paper here: [https://aclanthology.org/2024.lrec-main.1467.pdf](#).