Datasets:

riotu-lab
/

ARCADE-full

dataset_info:
  config_name: full_dataset
  features:
    - name: audio
      dtype: audio
    - name: filename
      dtype: string
    - name: duration
      dtype: string
    - name: country
      dtype: string
    - name: city
      dtype: string
    - name: msa_or_dialect
      dtype: string
    - name: emotion
      dtype: string
    - name: keep_or_skip
      dtype: string
    - name: confidence
      dtype: string
    - name: audio_type
      dtype: string
    - name: annotators
      dtype: string
    - name: timestamp
      dtype: string
  splits:
    - name: train
      num_bytes: 4191724917
      num_examples: 6907
  download_size: 3980038261
  dataset_size: 4191724917
configs:
  - config_name: full_dataset
    data_files:
      - split: train
        path: full_dataset/train-*
license: cc-by-4.0
task_categories:
  - audio-classification
language:
  - ar
tags:
  - arabic
  - dialect-identification
  - speech
  - radio
  - geolocation
pretty_name: ARCADE

ARCADE: Arabic Radio Corpus for Audio Dialect Evaluation

ARCADE is a city-scale corpus of Arabic radio speech designed for fine-grained dialect identification. The dataset contains 6,907 annotations for 3,790 unique audio segments collected from radio streams spanning 58 cities across 19 Arab countries.

Dataset Description

Each 30-second audio clip is annotated with:

City and Country: Fine-grained geographic labels at the city level
MSA or Dialect: Whether the speech is Modern Standard Arabic, dialectal, mixed, or not applicable
Emotion: Speaker emotion (neutral, happiness, anger, etc.)
Audio Type: Single speaker, multiple speakers, music/no speech, or Quran recitation
Keep or Skip: Whether the clip is suitable for dialect modeling
Confidence: Annotator confidence level (sure, unsure, no idea)

The filename alone is not a unique identifier and may appear across multiple cities. To obtain a unique key, concatenate the filename with the corresponding city name.

A detailed description of the dataset is provided in the accompanying paper: https://arxiv.org/abs/2601.02209

Intended Uses

Fine-grained Arabic dialect identification at the city level
Sociolinguistic studies of regional speech variation
Multi-task learning combining dialect, emotion, and speaker classification
Robustness evaluation under domain and channel shift

Dataset Statistics

Total annotations: 6,907
Total unique audio segments: 3,790
Cities: 58
Countries: 19
Clip duration: 30 seconds

Usage

from datasets import load_dataset
ds = load_dataset("riotu-lab/ARCADE-full")

Citation

If you use this dataset, please cite:

@misc{nacar2026arcadecityscalecorpusfinegrained,
      title={ARCADE: A City-Scale Corpus for Fine-Grained Arabic Dialect Tagging}, 
      author={Omer Nacar and Serry Sibaee and Adel Ammar and Yasser Alhabashi and Nadia Samer Sibai and Yara Farouk Ahmed and Ahmed Saud Alqusaiyer and Sulieman Mahmoud AlMahmoud and Abdulrhman Mamdoh Mukhaniq and Lubaba Raed and Sulaiman Mohammed Alatwah and Waad Nasser Alqahtani and Yousif Abdulmajeed Alnasser and Mohamed Aziz Khadraoui and Wadii Boulila},
      year={2026},
      eprint={2601.02209},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2601.02209}, 
}

📄 Paper: arXiv:2601.02209

License

This dataset is released under the CC BY 4.0 license for non-commercial academic and research use.