README.md

metadata

dataset_info:
  features:
    - name: audio
      dtype: audio
    - name: text
      dtype: string
    - name: cleaned_text
      dtype: string
    - name: speaker_age
      dtype: string
    - name: speaker_gender
      dtype: string
    - name: speaker_dialect
      dtype: string
  splits:
    - name: train
      num_bytes: 27052965505.604
      num_examples: 241834
    - name: validation
      num_bytes: 1015213004.222
      num_examples: 5139
    - name: test
      num_bytes: 701540013.966
      num_examples: 6193
  download_size: 50401167803
  dataset_size: 28769718523.792
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
      - split: test
        path: data/test-*
task_categories:
  - automatic-speech-recognition
  - text-to-speech
language:
  - ar
pretty_name: SADA 2022
size_categories:
  - 100K<n<1M
license: cc-by-nc-sa-4.0

Dataset Card for SADA (Saudi Audio Dataset for Arabic)

Dataset Summary

The SADA dataset (Saudi Audio Dataset for Arabic) is a large-scale Arabic speech corpus designed to support the development of high-quality artificial intelligence models for Arabic speech processing. It contains over 667 hours of transcribed Arabic audio recordings, primarily featuring various Saudi dialects, and was curated in a collaboration between the National Center for Artificial Intelligence at SDAIA and the Saudi Broadcasting Authority.

The dataset includes diverse spoken content extracted from more than 57 TV shows, encompassing a variety of speakers, dialects, and speech contexts. The corpus is accompanied by metadata including speaker age group, gender, and dialect, making it suitable for a wide range of speech and language modeling tasks.

Supported Tasks and Leaderboards

The dataset is suitable for training and evaluating models in:

• Automatic Speech Recognition (ASR)
• Text-to-Speech (TTS)
• Speaker Diarization
• Dialect Identification
• Gender and Age Classification

Languages

• Arabic (ar) — various regional dialects of Saudi Arabia (Najdi, Hijazi, Khaliji)

Dataset Structure

Data Fields

• audio: The raw audio recording in a supported format (e.g., .wav)
• text: The original transcription of the audio
• cleaned_text: A normalized version of the transcription
• speaker_age: Age group of the speaker (e.g. adult, elderly, or unknown)
• speaker_gender: Gender of the speaker (e.g. male, female, or unknown)
• speaker_dialect: Dialect classification (e.g. najidi, hijazi, khaliji, or unknown)

Splits

• Train: ~647 hours
• Test/Dev: ~20 hours

Dataset Creation

Curation

The data was sourced from publicly available TV content provided by the Saudi Broadcasting Authority (SBA) and manually transcribed by the National Center for Artificial Intelligence. The audio was processed, segmented, and annotated to ensure usability in machine learning applications.

Motivation

Due to the scarcity of open Arabic speech datasets, especially with dialectal variety, SADA aims to empower the research and development of Arabic-centric AI solutions. It enables advancements in speech technologies while promoting the Arabic language, which is spoken by over 400 million people globally and holds deep cultural and religious significance.

Licensing

This dataset is licensed under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0) license.

More details: https://creativecommons.org/licenses/by-nc-sa/4.0/

Citation

@misc{SADA2022,
  title={SADA: Saudi Audio Dataset for Arabic},
  author={SDAIA and Saudi Broadcasting Authority},
  year={2022},
  howpublished={\url{https://www.kaggle.com/datasets/sdaiancai/sada2022}},
  note={CC BY-NC-SA 4.0}
}

Contributions

Dataset curated by SDAIA in collaboration with SBA. Dataset card written by the community.