Dataset Card for NST-da

Dataset Summary

This dataset is an upload of the NST Danish ASR Database (16 kHz) – reorganized.

The training 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: 5.36 GB
• Size of the generated dataset: 64.09 GB
• Total amount of disk used: 69.45 GB

An example from the dataset looks as follows.

{
  'audio': {
    'path': 'dk14x404-05072000-1531_u0008121.wav',
    'array': array([ 0.00265503,  0.00248718,  0.00253296, ..., -0.00030518,
                    -0.00035095, -0.00064087]),
    'sampling_rate': 16000
  },
  'text': 'Desuden er der en svømmeprøve, en fremmedsprogstest samt en afsluttende samtale.',
  'speaker_id': 404,
  'age': 24,
  'sex': 'Female',
  'dialect': 'Storkøbenhavn',
  'recording_datetime': '2000-07-05T15:31:14'
}

Data Fields

The data fields are the same among all splits.

• audio: an Audio feature.
• text: a string feature.
• speaker_id: an int64 feature.
• age: an int64 feature.
• sex: a string feature.
• dialect: a string feature.
• recording_datetime: a string feature.

Dataset Statistics

There are 183,205 samples in the training split, and 54,747 samples in the test split.

Speakers

There are 539 unique speakers in the training dataset and 56 unique speakers in the test dataset, where 54 of them are also present in the training set.

Age Distribution

Dialect Distribution

Sex Distribution

Transcription Length Distribution

Dataset Creation

Curation Rationale

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

Source Data

The data originates from the now bankrupt company Nordisk språkteknologi (NST), whose data was transferred to the National Library of Norway, who subsequently released it into the public domain.

Additional Information

Dataset Curators

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 the CC0 license.