wikipedia_asr / README.md
rmndrnts's picture
Update README.md
c7591c8 verified
metadata
dataset_info:
  features:
    - name: audio
      dtype: audio
    - name: transcription
      dtype: string
    - name: speaker
      dtype: string
    - name: category
      dtype: string
  splits:
    - name: train
      num_bytes: 43917886678.336
      num_examples: 140482
  download_size: 39649854722
  dataset_size: 43917886678.336
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

Synthetic dataset based on highly specialised texts from Wikipedia. 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. Splitted by category version can be found here: https://huggingface.co/datasets/rmndrnts/wikipedia_asr_splitted