MohamedGomaa30/Masri-Elders

Masri Elders Speech-to-Text Dataset

Dataset Description

This dataset contains Egyptian Arabic (Masri) speech recordings from elderly speakers, paired with their transcriptions. It is designed to help improve Automatic Speech Recognition (ASR) models for this specific dialect and demographic, which is often underrepresented in standard datasets.

• Language: Egyptian Arabic (Masri)
• Demographic: Elders
• Sampling Rate: 16kHz (Mono)
• Format: WAV audio + Text transcripts

Dataset Structure

The dataset follows a standard audio dataset format:

• wavs/: Directory containing the audio segments.
• metadata.csv: A CSV file mapping audio files to transcripts.

Metadata Schema

Column	Description
file_name	Path to the audio file (e.g., wavs/vid_0_segment_0000.wav)
text	The transcribed text for the segment
start	Start time of the segment in the original audio (seconds)
end	End time of the segment in the original audio (seconds)
video_id	Identifier for the source video/audio

Usage

You can load this dataset directly using the Hugging Face datasets library:

from datasets import load_dataset

dataset = load_dataset("MohamedGomaa30/Masri-Elders")

# Listen to a sample
print(dataset['train'][0])

Source Code

The pipeline used to generate this dataset is open-source and available on GitHub: GitHub Repository: Mohamed-Gomaa30/Masri-Elders-Dataset

License

This dataset is released under the MIT License.