metadata

language:
  - ta
  - ms
  - en
configs:
  - config_name: default
    data_files:
      - split: tamil
        path: data/tamil-00000-of-00001.parquet
      - split: tamil_en
        path: data/tamil_en-00000-of-00001.parquet
      - split: tamil_ms
        path: data/tamil_ms-00000-of-00001.parquet

Pseudolabel Tamil using Whisper Large V3

Original dataset is newsonair_v5 from AI4Bharat, Heavy postprocessing and post-translation to improve pseudolabeled Whisper Large V3.

prepare dataset

wget https://gist.githubusercontent.com/huseinzol05/2e26de4f3b29d99e993b349864ab6c10/raw/9b2251f3ff958770215d70c8d82d311f82791b78/unzip.py
huggingface-cli download --repo-type dataset \
--include 'newsonair_v5-*.zip' \
--local-dir './' \
mesolitica/pseudolabel-tamil-large-v3-timestamp
python3 unzip.py