README.md

metadata

license: apache-2.0
language:
  - hi

CommonVoiceCorpusHindi15

Directory structure:

1. assets
   a. Download whole compressed dataset by clicking on the cv-corpus-15.0-2023-09-08-hi.tar.gz file.
   b. splitdata.py, python script contains code to split "clips" direcrtory in the original dataset. Because HuggingFace supports 10,000 files per directory but in the original dataset "clips" directory contains 14,000 files almost. So, "clips" directory is splitted into two directories "clips0" and "clips1". "clips0" contains 10,000 audio files exactly and "clips1" contains all remaining audio files.

2. data
   a. "clips0" contains 10,000 audio files
   b. "clips1" contains 4,000+ audio files (remaining from 14,000 audio files) c. Remaining all are metadata files.

3. Credit: Common Voice moz://a