rmndrnts's picture
Update README.md
bcdb3c7 verified
metadata
dataset_info:
  features:
    - name: audio
      dtype: audio
    - name: transcription
      dtype: string
    - name: speaker
      dtype: string
    - name: category
      dtype: string
  splits:
    - name: chemistry
      num_bytes: 5614555106.796
      num_examples: 17591
    - name: computergames
      num_bytes: 5610652921.056
      num_examples: 17484
    - name: cytologygenetics
      num_bytes: 4554410195.82
      num_examples: 12855
    - name: economics
      num_bytes: 5106377220.938
      num_examples: 14069
    - name: linguistics
      num_bytes: 4620074599.26
      num_examples: 12990
    - name: mathematics
      num_bytes: 3109788709.224
      num_examples: 11158
    - name: mlnlp
      num_bytes: 1463497003.336
      num_examples: 4063
    - name: philosophy
      num_bytes: 10148260747.051
      num_examples: 26417
    - name: politics
      num_bytes: 6389287932.055
      num_examples: 17985
    - name: zoology
      num_bytes: 1332449573.66
      num_examples: 5870
  download_size: 39652454717
  dataset_size: 47949354009.19601
configs:
  - config_name: default
    data_files:
      - split: chemistry
        path: data/chemistry-*
      - split: computergames
        path: data/computergames-*
      - split: cytologygenetics
        path: data/cytologygenetics-*
      - split: economics
        path: data/economics-*
      - split: linguistics
        path: data/linguistics-*
      - split: mathematics
        path: data/mathematics-*
      - split: mlnlp
        path: data/mlnlp-*
      - split: philosophy
        path: data/philosophy-*
      - split: politics
        path: data/politics-*
      - split: zoology
        path: data/zoology-*

Synthetic dataset based on highly specialised texts from Wikipedia. This version is splitted by category. Voiced using Yandex SpeechKit with random voices, roles and speech rate. Can be used to evaluate ASR models not trained on given domains and to identify areas that the model does not handle well. Non-splitted version can be found here: https://huggingface.co/datasets/rmndrnts/wikipedia_asr