Update README.md

README.md

@@ -1,3 +1,22 @@
----
-license: cc-by-sa-4.0
----
+---
+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.