README.md

metadata

dataset_info:
  config_name: hi
  features:
    - name: path
      dtype: string
    - name: audio
      dtype:
        audio:
          sampling_rate: 16000
    - name: sentence
      dtype: string
    - name: condition_on_prev
      sequence: int64
    - name: whisper_transcript
      dtype: string
  splits:
    - name: train
      num_bytes: 634978222
      num_examples: 717
    - name: validation
      num_bytes: 359634707
      num_examples: 405
    - name: test
      num_bytes: 501454543
      num_examples: 575
  download_size: 1400598630
  dataset_size: 1496067472
configs:
  - config_name: hi
    data_files:
      - split: train
        path: hi/train-*
      - split: validation
        path: hi/validation-*
      - split: test
        path: hi/test-*

Common Voice 16.1 Hindi Pseudo-Labelled

This is the Common Voice 16.1 Hindi split pseudo-labelled using the Whisper large-v3 model, according to the instructions detailed in the Distil-Whisper repository. To reproduce this pseudo-labelling run, follow the instructions detailed here.