README.md

metadata

dataset_info:
  - config_name: full
    features:
      - name: audio
        dtype: audio
      - name: file_number
        dtype: string
      - name: show_title
        dtype: string
      - name: episode_title
        dtype: string
      - name: itunes_id
        dtype: string
      - name: transcription
        dtype: string
    splits:
      - name: test
        num_bytes: 1509910660
        num_examples: 30
    download_size: 1445493754
    dataset_size: 1509910660
  - config_name: whisper_subset
    features:
      - name: audio
        dtype: audio
      - name: file_number
        dtype: string
      - name: show_title
        dtype: string
      - name: episode_title
        dtype: string
      - name: itunes_id
        dtype: string
      - name: transcription
        dtype: string
    splits:
      - name: test
        num_bytes: 921693242
        num_examples: 16
    download_size: 881542397
    dataset_size: 921693242
configs:
  - config_name: full
    data_files:
      - split: test
        path: full/test-*
  - config_name: whisper_subset
    data_files:
      - split: test
        path: whisper_subset/test-*

Dataset Card for "rev16"

Configs:

• full: the entire 30 podcast files
• whisper_subset: the subset of 16 podcast files used in the Whisper paper for long-form evaluation. The remaining 14 files have mis-matches between the audio and labels, and are thus filtered from the test set.