projecte-aina/parlament_parla_v3

Dataset Card for ParlamentParla v3 - Speech Corpus of Catalan Parliamentary Sessions

A speech corpus composed of Catalan Parliamentary Sessions.The v3 and last version of the corpus includes both clean and other quality segments, divided into short segments (less than 30 seconds) and long segments (more than 30 seconds). The total dataset encompasses 1059h 48m 04s of speech, including 945h 51m 06s for the short segments and 113h 56m 58s for the long segments, with a total of 10.925.943 words.

Dataset Details

Dataset Description

This is the third version of the ParlamentParla speech corpus for Catalan: a collection of speech recordings with transcriptions intended for Automatic Speech Recognition (ASR) applications. In recent years, data in Catalan language has increased considerably. With this release we develop the third version of the ParlamentParla speech corpus, which will be very valuable mainly for training and evaluating speech recognition systems. We used the content of the Catalan Parlamentary sessions: the audio segments were extracted from recordings the Catalan Parliament (Parlament de Catalunya) plenary sessions. Taking advantage of the manual transcriptions, we created high quality audio segments in Catalan along with the aligned transcriptions. The extensive time span covered by the sessions, from January 23, 2008, to October 24, 2023, provides a broad range of linguistic phenomena and topics, further enriching the corpus. With 317 sessions in total, the corpus is substantial and should provide ample data for various research and development purposes in speech recognition. The final corpus has been extracted March 6, 2024.

• Curated by: Language Technologies Unit at the Barcelona Supercomputing Center (langtech@bsc.es)
• Funded by: This work/research has been promoted and financed by the Government of Catalonia through the Aina project.
• Shared by: [More Information Needed]
• Language(s) (NLP): ca-ce
• License: CC-BY-4.0

Dataset Sources

• Repository:
• Paper:

Uses

The purpose of this dataset is mainly for training automatic speech recognition (ASR) models in Catalan. The language is limited to the plenary sessions of the parlament used to create the corpus and may not be representative to all domains.

Dataset Structure

Data Instances

Each instance have the following structure:

DatasetDict({
    clean_train: Dataset({
        features: ['identifier','audio','segment_path','text'],
        num_rows: 164416
    })

Each data point is structured as:

• Audio ID

>>data['clean_train_short'][0]['identifier']

2753976_90753a8d81888d998484_405.96_411.15999999999997

• Audio

>>data['clean_train_short'][0]['audio']

{'path': '/Users/sarahsolito/.cache/huggingface/datasets/downloads/extracted/9f760c175adf0af8127242f9468e48120f7682b20cf5c5813bfe481a108524bf/parlament_parla_v3/corpus/speech/2753976/2753976_90753a8d81888d998484_405.96_411.15999999999997.wav', 'array': array([-1.07421875e-02, -1.33972168e-02, -1.62353516e-02, ...,
        1.64794922e-03,  3.05175781e-05, -4.02832031e-03]), 'sampling_rate': 16000}

• Relative Path

>>data['clean_train_short'][0]['segment_path']

data/parlament_parla_v3/output_segment/2753976/2753976_90753a8d81888d998484_405.96_411.15999999999997.wav

• Transcription

>>data['clean_train_short'][0]['text'])

idò jo em tragaré el salmó oh uh no hi pensava

Data Fields

• "identifier" : (string) → the unique audio identificator
• "segment_path": (string) → the path to the audio
• "audio": datasets.Audio(sampling_rate=16000) → the decoded audio array, and the sampling rate.
• "text": (string) → clean version of the transcription

Data Splits

The dataset consists of a train, dev and test splits. The stat details are as follows:

Subcorpus	Duration
other_test_short	13:42:44
other_dev_short	13:13:45
other_train_short	507:27:34
other total_short	534:24:03
clean_test_short	10:44:19
clean_dev_short	10:23:30
clean_train_short	390:19:12
clean total_short	411:27:03
Total	945:51:06

Subcorpus	Duration
other_test_long	01:41:29
other_dev_long	01:51:30
other_train_long	72:35:10
other total_long	76:08:10
clean_test_long	00:50:15
clean_dev_long	00:46:44
clean_train_long	36:11:46
clean total_long	37:48:47
Total	113:56:58

:04

Example Usage

To load a specific split ,for example, the training split do:

  from datasets import load_dataset
  data = load_dataset("projecte-aina/parlament_parla_v3",split="clean_train_short")

Dataset Creation

Curation Rationale

The directory called "speech" contains all the speech files of the corpus. The files in the speech directory are divided into the "clean" and the "other" directories.

Source Data

The content belongs to the Catalan Parliament and the data is released conforming their terms of use.

Data Collection and Processing

The dataset's transcriptions are released in a clean version. The clean versions have been normalized at an orthographic level in lower-case. The normalization process was performed removing punctuation marks and characters that are not present in the Catalan alphabet. Number expansion was also perfomed.

In order to obtain a corpus of the highest possible quality, we also apply automatic language detection processes to each segment to prevent code-switching, and evaluate the quality of the transcriptions to eliminate both low quality segments and those that are not in Catalan.

Who are the source data producers?

The content belongs to the Catalan Parliament and the data is released conforming their terms of use.

Annotations

The dataset doesn't contain any additional annotation.

Personal and Sensitive Information

The dataset consists of Catalan parliamentary speeches and their transcription. The dataset contains no personal information except for speech, which is considered personal data. Consequently, the speakers' voices in this corpus have been subjected to anonymization treatment in compliance with applicable regulations, such as the General Data Protection Regulation (GDPR) in the European Union. You agree to not attempt to determine the identity of speakers in this dataset.

Citation

@misc{bscib32024,
      title={ParlamentParla v3 - Speech Corpus of Catalan Parliamentary Sessions}, 
      author={Baybars, Kulebi},
      publisher={Barcelona Supercomputing Center},
      year={2024},
      url={},
      }

Considerations for Using the Data

Social Impact of Dataset

ParlamentParla_v3 is a source of speech data that will be valuable in development of speech technologies for Catalan language and its varieties.

Discussion of Biases

No specific bias mitigation strategies were applied to this dataset. Inherent biases may exist within the data.

Other Known Limitations

Speakers, their gender and age are not identified and one or more speakers could be speaking in the same recording. For these reasons, we don't know the total number of speakers in the corpus and their gender/age.