---
license: cc-by-sa-4.0
language:
- ar
task_categories:
- automatic-speech-recognition
---

# Wikitongues-Darija

This is a small test dataset for Automatic Speech Recognition in Darija language, built from 2 captioned videos of the WikiTongues project:

- [nawal](https://en.wikipedia.org/wiki/File:WIKITONGUES-_Nawal_speaking_Moroccan_Arabic.webm)
- [anass](https://en.wikipedia.org/wiki/File:WIKITONGUES-_Anass_speaking_Moroccan_Arabic.webm)

Process:

- each webm video has been converted to monochannel 16khz wav files with ffmpeg :
```bash
ffmpeg -i WIKITONGUES-_Nawal_speaking_Moroccan_Arabic.webm.1080p.vp9.webm -ar 16000 -ac 1 nawal.wav
```
- each audio has been cut in samples of less than 30 seconds audio according to the captions timestamps. The script may be found at the root of the dataset.