alexandrainst/ftspeech

Dataset Card for FT Speech

Dataset Summary

This dataset is an upload of the FT Speech dataset.

The training, validation and test splits are the original ones.

Supported Tasks and Leaderboards

Training automatic speech recognition is the intended task for this dataset. No leaderboard is active at this point.

Languages

The dataset is available in Danish (da).

Dataset Structure

Data Instances

• Size of downloaded dataset files: 101.78 GB
• Size of the generated dataset: 214.15 GB
• Total amount of disk used: 315.93 GB

An example from the dataset looks as follows.

{
  'utterance_id': 'S001_20151_M012_P00034-2',
  'speaker_gender': 'F',
  'sentence': 'alle de fem tekniske justeringer der er en del af lovforslaget',
  'speaker_id': 'S001',
  'audio': {
    'path': 'S001_20151_M012_P00034-2.wav',
    'array': array([-3.75366211e-03, -5.27954102e-03, -3.87573242e-03, ...,
                    9.15527344e-05, -1.52587891e-04,  5.79833984e-04]),
    'sampling_rate': 16000
  }
}

Data Fields

The data fields are the same among all splits.

• utterance_id: a string feature.
• speaker_gender: a string feature.
• sentence: a string feature.
• speaker_id: a string feature.
• audio: an Audio feature.

Dataset Statistics

There are 995,677 samples in the training split, 2,601 in the dev_balanced split, 7,595 in the dev_other split, 5,534 in the test_balanced and 5,837 in the test_other split.

Speakers

There are 374 unique speakers in the training dataset, 20 unique speakers in the validation dataset and 40 unique speakers in the test dataset. None of the dataset splits share any speakers.

Gender Distribution

Transcription Length Distribution

Dataset Creation

Curation Rationale

There are not many large-scale ASR datasets in Danish.

Source Data

The data constitutes public recordings of sessions from the Danish Parliament, along with manual transcriptions.

Additional Information

Dataset Curators

Andreas Kirkedal, Marija Stepanović and Barbara Plank curated the dataset as part of their FT Speech paper (see citation below).

Dan Saattrup Nielsen from the The Alexandra Institute reorganised the dataset and uploaded it to the Hugging Face Hub.

Licensing Information

The dataset is licensed under this custom license.

Citation

@inproceedings{ftspeech,
  author     = {Kirkedal, Andreas and Stepanović, Marija and Plank, Barbara},
  title      = {{FT Speech: Danish Parliament Speech Corpus}},
  booktitle  = {Proc. Interspeech 2020},
  year       = {2020},
  url        = {arxiv.org/abs/2005.12368}
}