metadata

YAML tags: null
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 Parliament Speech Corpus (NPSC) is a corpus for training a Norwegian ASR (Automatic Speech Recognition) models. The corpus is created by Språkbanken at the National Library in Norway.

NPSC is based on sound recording from meeting in the Norwegian Parliament. These talks are orthographically transcribed to either Norwegian Bokmål or Norwegian Nynorsk. There are a significant amount of metadata related to the speakers, as well as minutes from the actual meetings.

The corpus is in total sound recordings from 40 entire days of meetings. This amounts to 140 hours of speech, 65,000 sentences or 1.2 million words.

This corpus is an adaption of the original corpus made for efficiant ASR training. For simplicity and portability, a few of the original datasets features, like the token transcription, is ommitted. You can find the complete dataset at the Resource Catalogue at Språkbanken.

How to Use (This needs to be edited of course)

from datasets import load_dataset
data = load_dataset("nb/NPSC", 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": [.......]
}

}

Data Fields

Currently there are two versions included in this repo.

Version A

This verison has a short list of the metadata and includes the audio (48k mp3) encoded as a float32 array in the dataset itself.

The current dataloader script is associated with this version.

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
text	String sentence text
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)

Version B

This verison does not contain the audio encoded in the dataset. Instead it has the audio files placed in sub-directories. There are currently both samples in clips_48k_wav and clips_16k_mp3. Only the base filename is referred in the dataset. Please not that there are both sentence-based audio clips as well at meeting-based audio clips. The dataset contains referrals to both, the latter referral has start and stop time as well.

Dataset Creation

We are providing a train, dev and test split. These are the same as in the orginal corpus.

Build date: 20012022

Initial Data Collection and Curation

The procedure for the dataset creation is described in detail in the 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%

Considerations for Using the Data

This corpus contains speech data and is allowed to be used outside the National Library of Norway for speech recognition technology purposes.

Discussion of Biases

Please refer to our paper.

Dataset Curators

Per Erik Solberg

Freddy Wetjen, Andre Kaasen and [Per Egil Kummervold](mailto: [email protected]) has contributed to porting it to Hugging Face.

Licensing Information

Licensed for use outside the National Library of Norway.

License

CC-ZERO(https://creativecommons.org/publicdomain/zero/1.0/)

Citation Information

We are preparing an article with detailed information about this corpus. Until it is published, please cite out paper discussing the first version of this corpus:

ANDRE: TO BE DONE

Datasets:

NbAiLab
/

NPSC_test