CoRal-project/coral-tts

Dataset Card for CoRal TTS

Dataset Summary

This dataset consists of two professional Danish speakers, female and male, recording roughly 17 hours of Danish speech each.

The dataset is part of the CoRal project which is funded by the Danish Innovation Fund.

The text data was selected by the Alexandra Institute (Github repo for the dataset creation) and consists of sentences from sundhed.dk, borger.dk, names of bus stops and stations, manually filtered Reddit comments, and dates and times.

The audio data was recorded by the public institution Nota, which is part of the Danish Ministry of Culture.

Supported Tasks and Leaderboards

Speech synthesis is the intended tasks 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: 14.63 GB
• Size of the generated dataset: 15.25 GB
• Total amount of disk used: 29.88 GB

An example from the dataset looks as follows.

{
 'speaker_id': 'mic',
 'transcription_id': 0,
 'text': '26 rigtige.',
 'audio': {
  'path': 'mic_00001.wav',
  'array': array([-0.00054932, -0.00054932, -0.00061035, ...,  0.00027466,
                   0.00036621,  0.00030518]),
  'sampling_rate': 44100
 }
}

Data Fields

The data fields are the same among all splits.

• speaker_id: a string feature.
• transcription_id: an int feature.
• text: a string feature.
• audio: an Audio feature.

Dataset Statistics

There are 18,863 samples in the dataset.

Additional Information

Dataset Curators

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

Licensing Information

The dataset is licensed under the CC0 license.