pere's picture
Upload dataset
c58e4bc verified
metadata
annotations_creators:
  - no-annotation
language_creators:
  - found
license:
  - other
multilinguality:
  - multilingual
pretty_name: ncc_speech_v7
size_categories:
  - 2G<n<1B
source_datasets:
  - original
task_categories:
  - automatic-speech-recognition
task_ids:
  - language-modeling
configs:
  - config_name: None
    description: This dataset does not need any config file.
  - config_name: 'no'
    data_files:
      - split: validation
        path: no/validation-*
      - split: test
        path: no/test-*
      - split: validation_audio_books_no
        path: no/validation_audio_books_no-*
      - split: validation_clean_audio_books_no
        path: no/validation_clean_audio_books_no-*
      - split: validation_clean_stortinget_no
        path: no/validation_clean_stortinget_no-*
      - split: validation_norwegian_fleurs
        path: no/validation_norwegian_fleurs-*
      - split: validation_nrk_no
        path: no/validation_nrk_no-*
      - split: validation_nst
        path: no/validation_nst-*
      - split: validation_stortinget_no
        path: no/validation_stortinget_no-*
      - split: test_audio_books_no
        path: no/test_audio_books_no-*
      - split: test_clean_audio_books_no
        path: no/test_clean_audio_books_no-*
      - split: test_clean_stortinget_no
        path: no/test_clean_stortinget_no-*
      - split: test_norwegian_fleurs
        path: no/test_norwegian_fleurs-*
      - split: test_nrk_no
        path: no/test_nrk_no-*
      - split: test_nst
        path: no/test_nst-*
      - split: test_stortinget_no
        path: no/test_stortinget_no-*
dataset_info:
  config_name: 'no'
  features:
    - name: id
      dtype: string
    - name: audio
      dtype:
        audio:
          sampling_rate: 16000
    - name: text
      dtype: string
    - name: condition_on_prev
      sequence: int64
    - name: whisper_transcript
      dtype: string
  splits:
    - name: validation
      num_bytes: 2454633826.882
      num_examples: 3342
    - name: test
      num_bytes: 2414144009.228
      num_examples: 3271
    - name: validation_audio_books_no
      num_bytes: 428451732
      num_examples: 577
    - name: validation_clean_audio_books_no
      num_bytes: 364866667
      num_examples: 490
    - name: validation_clean_stortinget_no
      num_bytes: 580737498
      num_examples: 694
    - name: validation_norwegian_fleurs
      num_bytes: 66508760
      num_examples: 103
    - name: validation_nrk_no
      num_bytes: 296038955
      num_examples: 396
    - name: validation_nst
      num_bytes: 309550647
      num_examples: 378
    - name: validation_stortinget_no
      num_bytes: 1277552109.67
      num_examples: 1541
    - name: test_audio_books_no
      num_bytes: 438716189
      num_examples: 598
    - name: test_clean_audio_books_no
      num_bytes: 373349469
      num_examples: 520
    - name: test_clean_stortinget_no
      num_bytes: 536931906
      num_examples: 647
    - name: test_norwegian_fleurs
      num_bytes: 143531633
      num_examples: 202
    - name: test_nrk_no
      num_bytes: 279512472
      num_examples: 376
    - name: test_nst
      num_bytes: 308603638
      num_examples: 382
    - name: test_stortinget_no
      num_bytes: 1159354882.49
      num_examples: 1395
  download_size: 11015262505
  dataset_size: 11432484394.27

Dataset Card: NbAiLab/distil_raw_ncc_speech_v7

  • Internal dataset created as input for creating Pseudo Labels.

General Information

The dataset is based on ncc_speech_v7 (Norwegian Colossal Corpus - Speech). It is then filtered by only including entries where the text language in Norwegian, and where the source is not from "nrk_translate".

Potential Use Cases

The ncc_speech_v7 corpus can be used for various purposes, including but not limited to:

  • Training Automatic Speech Recognition models.
  • Building text-to-speech systems.
  • Research in speech recognition and natural language processing.
  • Developing language models.

License

The ncc_speech_v7 corpus has a private license.

Citation

The corpus was created and cleaned by Freddy Wetjen, Rolv-Arild Braaten, Angelina Zanardi and Per Egil Kummervold. No publication is so far published based on this copus.