DACS: Dialectal Arabic Code-Switching Dataset

This release includes 2 hours of annotated Egyptian Arabic speech from the ADI-5 development split in the MGB-3 Challenge.

Overview

Unlike the original MGB-3:EGY data, this corpus has been manually segmented into smaller utterances based on pauses of ≥500 ms of silence.
Each utterance has been verbatim transcribed by a lay native Egyptian speaker.

The released MGB-3 data includes both speech features and textual features extracted from ASR transcriptions.

Download

📦 Dataset Access
You can download the dataset from the following link:
👉 Download DACS Dataset

By downloading, you agree to use the data for non-commercial research purposes only.

Annotation Details

The transcribed data was further annotated for word-level Code-Switching (CS) by three independent annotators.
Following the guidelines described in the associated paper, each word was classified into one of the following four categories:

• MSA — Modern Standard Arabic word with MSA pronunciation
• EGY — Egyptian dialect word
• MIX — MSA word with dialectal pronunciation
• FRN — Foreign word (i.e., non-Arabic)

An additional tag NULL was used for unintelligible words or those that could not be categorized into any of the four labels.

Citation

If you use this dataset, please cite the following paper:

@inproceedings{chowdhury2020effects,
  title={Effects of Dialectal Code-Switching on Speech Modules: A Study using Egyptian Arabic Broadcast Speech},
  author={Chowdhury, Shammur Absar and Samih, Younes and Eldesouki, Mohamed and Ali, Ahmed},
  booktitle={{{Proc. of INTERSPEECH}}},
  year={2020}
}