metadata

annotations_creators:
  - no-annotation
language_creators:
  - found
languages:
  - nb,no,nn
licenses:
  - CC-ZERO
multilinguality:
  - monolingual
pretty_name: NPSC
size_categories:
  - 2G<n<1B
source_datasets:
  - original
task_categories:
  - sequence-modeling
task_ids:
  - speech-modeling

Dataset Card for NbAiLab/NPSC

Dataset Description
Dataset Summary
Data Fields
Dataset Creation
Statistics
Document Types
Languages
Publish Periode
Considerations for Using the Data
Social Impact of Dataset
Discussion of Biases
Other Known Limitations
Additional Information
Dataset Curators
Licensing Information
Citation Information

Dataset Description

Homepage: https://www.nb.no/sprakbanken/
Repository: https://www.nb.no/sprakbanken/ressurskatalog/oai-nb-no-sbr-58/
Paper: https://www.nb.no/sprakbanken/
Point of Contact: Per Erik Solberg

The Norwegian Parliamentary Speech Corpus (NPSC) is a speech corpus made by the Norwegian Language Bank at the National Library of Norway in 2019-2021. The NPSC consists of recordings of speech from Stortinget, the Norwegian parliament, and corresponding orthographic transcriptions to Norwegian Bokmål and Norwegian Nynorsk. All transcriptions are done manually by trained linguists or philologists, and the manual transcriptions are subsequently proofread to ensure consistency and accuracy. Entire days of Parliamentary meetings are transcribed in the dataset.

This repository contains a version of the NPSC tailored for Huggingface. Note that the official release of the dataset, which can be found in the repository of the Norwegian Language Bank, contains more information that the version found here, including word-level metadata, metadata about the speakers, and detailed documentation.

How to Use

# Loads the 16K Bokmål corpus in streaming mode
from datasets import load_dataset
data = load_dataset("NbAiLab/NPSC", config="16K_mp3_bokmaal", streaming=True)

Dataset Summary

The NPSC dataset contains json lines with language training data. Here is an example json line:


{
"sentence_id": 49853,
"sentence_order": 0,
"speaker_id": 32,
"speaker_name": "Olemic Thommessen",
"sentence_text": "Stortingets møte er lovlig satt",
"sentence_language_code": "nb-NO",
"text": "Stortingets møte er lovlig satt",
"start_time": 320246, "end_time": 323590,
"normsentence_text": "Stortingets møte er lovlig satt",
"transsentence_text": "Stortingets møte er lovleg sett",
"translated": 1,
"audio": {"path": "audio/20170110-095504_320246_323590.wav",
"array": [.......]
}

}
!!!PES: Not all the fields above and below correspond to what is actually found
in the json files. It would be good if you could correct this. When you have, I can
add further descriptions of the fields where needed.!!!

Data Fields (!!!!We really need a descrption here - and maybe some cleaning!!!)

id:	String with id to source of line and a unique identifier
sentence_order	String with order of sentence
speaker id	Integer id of speaker
speaker_name	String name of speaker
sentence_text	String sentence text
sentence_language_code	String sentence text (!!!LIST ALL ALTERNATIVES!!!)
text	String sentence text. This is a copy of "sentence_text". It is included here to make it more convenient to interleave with other datasets.
start_time	int start time
end_time	int end time
normsentence_text	String normalised sentence text
transsentence_text	String translated sentence text
translated	int text translated
audio	audio audio record with 'path',(mp3) 'array','sampling_rate' (48000)

Dataset Creation

We are providing a train and a validation split. The standard size of the validation is a single 1GB file, while train is sharded in 1GB chunks. All files are gzipped. There is also a test split, which is not shared here. It is, however part of the official release of the NPSC.

Initial Data Collection

The procedure for the dataset creation is described in detail in our paper.

Statistics

Feature	Value
Duration, pauses included	140,3 hours
Duration, pauses not included	125,7 hours
Word count	1,2 million
Sentence count	64.531
Language distribution	Nynorsk: 12,8%
	Bokmål: 87,2%
Gender distribution	Female: 38,3%
	Male: 61.7%

!!!It would be great to know how many hours of Bokmål/Nynorsk/English(?) here!!!!! !!!PES: As long as we have the percentages, I don't think it is necessary!!!

Considerations for Using the Data

This corpus contains speech data. All recordings are of members of Parliament in a public setting, and can be distributed without any restrains.

Dataset Creators and Curators

The content of the dataset was created by the Norwegian Language Bank (Språkbanken) at the National Library of Norway. Javier de la Rosa, Freddy Wetjen, Per Egil Kummervold, and Andre Kaasen all contributed in making this into a HuggingFace Dataset. Thanks to the HuggingFace team for assistance.

License

The sound and the transcriptions are released under the CC-ZERO-license. The curation of the HuggingFace Dataset is released under CC-BY-SA-3-license.

Citation Information

The following article gives detailed information about the corpus. Please refer to the article and this page if you are using this dataset:


@misc{solberg2022norwegian,
      title={The Norwegian Parliamentary Speech Corpus},
      author={Per Erik Solberg and Pablo Ortiz},
      year={2022},
      eprint={2201.10881},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

Datasets:

NbAiLab
/

NPSC