---
dataset_info:
  features:
    - name: audio
      dtype:
        audio:
          sampling_rate: 16000
    - name: sentence
  splits:
   - name: train
     num_bytes: 1094421637.73819888
     num_examples: 1138
  download_size: 888280535
  dataset_size: 1094421637.7381988
language:
- ar
tags:
- arabic
- egypt
- egyptian
- ASR
- automatic speech recognition
pretty_name: 'Egyptian Arabic dialect automatic speech recognition '
size_categories:
- 1K<n<10K
task_categories:
- automatic-speech-recognition
---
# Egyptian Arabic dialect automatic speech recognition 

## Dataset Description

- **Homepage:** 
- **Repository:** 
- **Paper:** 
- **Leaderboard:** 
- **Point of Contact:** 

### Dataset Summary

This dataset was collected, cleaned and adjusted for huggingface hub and ready to be used for whisper finetunning/training.

[From MGB-3 website](http://www.mgb-challenge.org/MGB-3.html):

*The MGB-3 is using 16 hours multi-genre data collected from different YouTube channels. The 16 hours have been manually transcribed. 
The chosen Arabic dialect for this year is Egyptian. 
Given that dialectal Arabic has no orthographic rules, each program has been transcribed by four different transcribers using this transcription guidelines.*

### Supported Tasks and Leaderboards

ASR: automatic speech recognition

### Languages

Arabic - Egyptian dialect

### Data Fields

* audio: sampled in 16000HZ and have a max duration of 30 sec (ideal for whispear and others ASR models)
* sentence: the transcription in Egyptian Arabic

## Dataset Creation

The youtube videos that are still avalible (some of them got deleted/ made private) were downloaded and synced with the provided transcription.
Then the 12 min of each youtube video were cut down into 30 sec segments.
the resulting dataset was uploaded to huggingface.

[From MGB-3 website](http://www.mgb-challenge.org/MGB-3.html):

*Egyptian broadcast data collected from YouTube.This year, we collected about 80 programs from different YouTube channels. The first 12 minutes from each program has been transcribed and released. This sums up to roughly 16 hours in total*

### Source Data

* [http://www.mgb-challenge.org/MGB-3.html](MGB challenge website)
* [Youtube.com](youtube)


#### Initial Data Collection and Normalization

[Available on MGB website ](http://www.mgb-challenge.org/MGB-3.html):

#### Who are the source language producers?

[Available on MGB website ](http://www.mgb-challenge.org/MGB-3.html):

### Annotations

#### Annotation process

[Available on MGB website ](http://www.mgb-challenge.org/MGB-3.html):

#### Who are the annotators?

[Available on MGB website ](http://www.mgb-challenge.org/MGB-3.html):

### Personal and Sensitive Information

[Available on MGB website ](http://www.mgb-challenge.org/MGB-3.html):


### Social Impact of Dataset

[Available on MGB website ](http://www.mgb-challenge.org/MGB-3.html):

### Discussion of Biases

[Available on MGB website ](http://www.mgb-challenge.org/MGB-3.html):

### Other Known Limitations

[Available on MGB website ](http://www.mgb-challenge.org/MGB-3.html):

## Additional Information

### Dataset Curators

[Available on MGB website ](http://www.mgb-challenge.org/MGB-3.html):

### Licensing Information

[Available on MGB website ](http://www.mgb-challenge.org/MGB-3.html):

### Citation Information

[Available on MGB website ](http://www.mgb-challenge.org/MGB-3.html)


[Speech Recognition Challenge in the Wild: Arabic MGB-3](https://arxiv.org/abs/1709.07276)


### Contributions