---
license: mit
language:
- ar
dataset_info:
  features:
  - name: audio
    dtype: audio
  - name: dialect
    dtype: string
  splits:
  - name: train
    num_bytes: 18915430866.025
    num_examples: 67015
  - name: validation
    num_bytes: 2907285972.342
    num_examples: 10806
  download_size: 21768558127
  dataset_size: 21822716838.367
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
---

# ADI-20 Micro

This is a smaller version of the ADI-20/ADI-17 Arabic dialect identification dataset use for the NADI 2026 shared task (although participants are encouraged to use the full ADI-20 for their submissions). This dataset consists of 10 hours per each dialect taken from the original ADI-17 alongside 10 hours for each of the new dialects introduced in ADI-20.




### Dataset Sources

<!-- Provide the basic links for the dataset. -->
- **ADI-17:** ArabicSpeech/ADI17

- **ADI-20:** ArabicSpeech/ADI20

- **Papers:** [ADI-17](https://ieeexplore.ieee.org/document/9003960), [ADI-20](https://www.isca-archive.org/interspeech_2025/elleuch25_interspeech.html)

## Uses

<!-- Address questions around how the dataset is intended to be used. -->

This dataset is designed for the training of Arabic dialect identification models and covers the following dialects:

'MSA', 'BAH', 'TUN', 'ALG', 'EGY', 'IRA', 'JOR', 'KSA', 'KUW', 'LEB', 'LIB', 'MAU', 'MOR', 'OMA', 'PAL', 'QAT', 'SUD', 'SYR', 'UAE', 'YEM'