README.md

metadata

dataset_info:
  features:
    - name: id
      dtype: string
    - name: audio
      dtype:
        audio:
          sampling_rate: 16000
    - name: text_indo
      dtype: string
    - name: text_en
      dtype: string
  splits:
    - name: train
      num_bytes: 35886292.768
      num_examples: 1243
    - name: validation
      num_bytes: 24899653
      num_examples: 792
    - name: test
      num_bytes: 26407823
      num_examples: 844
  download_size: 85945718
  dataset_size: 87193768.768
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*
language:
  - id
  - en
pretty_name: d

Dataset Details

This is the Indonesia-to-English dataset for Speech Translation tasks. This dataset is acquired from CoVoST2. CoVoST2 is a corpus that is intended for speech-to-text translation tasks. CoVoST2 consists of 21 languages that are translated into English, one of them is Indonesian. This dataset approximately has 3 hours 6 minutes 58 seconds of audio data.

Processing Steps

Before the data is extracted, there are some preprocessing steps to the data:

1. Checked the duplicate ids in each splits.
2. Checked the overlap ids in across of the splits.
3. Removed some of the columns, except ids, Indonesian audio, sentence, and translation.
4. Renamed column "sentence" into "text_indo" and column "translation" into "text_en".

Dataset Structure

DatasetDict({
    train: Dataset({
        features: ['id', 'audio', 'text_indo', 'text_en'],
        num_rows: 1243
    }),
    validation: Dataset({
      features: ['id', 'audio', 'text_indo', 'text_en'],
      num_rows: 792
    }),
    test: Dataset({
      features: ['id', 'audio', 'text_indo', 'text_en'],
      num_rows: 844
    })
  
})

Citation

@misc{wang2020covost,
    title={CoVoST 2: A Massively Multilingual Speech-to-Text Translation Corpus},
    author={Changhan Wang and Anne Wu and Juan Pino},
    year={2020},
    eprint={2007.10310},
    archivePrefix={arXiv},
    primaryClass={cs.CL}

Credits:

Huge thanks to Yasmin Moslem for mentoring me.