README.md

metadata

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
• anass

Process:

• each webm video has been converted to monochannel 16khz wav files with ffmpeg :

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.