metadata

dataset_info:
  features:
    - name: audio
      dtype: audio
    - name: transcription
      dtype: string
  splits:
    - name: train
      num_bytes: 11189910118.05
      num_examples: 50715
    - name: validation
      num_bytes: 385055065.35
      num_examples: 2199
  download_size: 11017987865
  dataset_size: 11574965183.4
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: validation
        path: data/validation-*
tags:
  - masrispeech
  - egyptian-arabic
  - arabic
  - speech
  - audio
  - asr
  - automatic-speech-recognition
  - speech-to-text
  - stt
  - dialectal-arabic
  - egypt
  - native-speakers
  - spoken-arabic
  - egyptian-dialect
  - arabic-dialect
  - audio-dataset
  - language-resources
  - low-resource-language
  - phonetics
  - speech-corpus
  - voice
  - transcription
  - linguistic-data
  - machine-learning
  - natural-language-processing
  - nlp
  - huggingface
  - open-dataset
  - labeled-data
task_categories:
  - automatic-speech-recognition
  - audio-classification
  - audio-to-audio
language:
  - arz
  - ar
pretty_name: MasriSpeech-Full

🗣️ MasriSpeech-Full: Large-Scale Egyptian Arabic Speech Corpus

MasriSpeech-Full Dataset Overview

🌍 Overview

MasriSpeech-Full is the largest open-source Egyptian Arabic (Masri) speech dataset, designed to advance Automatic Speech Recognition (ASR) and speech processing research for dialectal Arabic. This corpus contains 52,914 professionally annotated audio samples totaling over 3,100 hours of natural Egyptian Arabic speech.

💡 Key Features:

High-quality 16kHz speech recordings

Natural conversational Egyptian Arabic

Speaker-balanced train/validation splits

Comprehensive linguistic coverage

Apache 2.0 license

📊 Dataset Summary

Feature	Value
Total Samples	52,914
Train Samples	50,715
Validation Samples	2,199
Sampling Rate	16 kHz
Total Duration	~3,100 hours
Languages	Egyptian Arabic (arz), Arabic (ar)
Format	Parquet
Dataset Size	11.57 GB
Download Size	10.26 GB
Annotations	Transcripts

🧱 Dataset Structure

The dataset follows Hugging Face datasets format with two splits:

DatasetDict({
    train: Dataset({
        features: ['audio', 'transcription'],
        num_rows: 50715
    })
    validation: Dataset({
        features: ['audio', 'transcription'],
        num_rows: 2199
    })
})

Data Fields

audio: Audio feature object containing:
- Array: Raw speech waveform (1D float array)
- Path: Relative audio path
- Sampling_rate: 16,000 Hz
transcription: string with Egyptian Arabic transcription

📈 Data Statistics

Split Distribution

Split	Examples	Size (GB)	Avg. Words	Empty	Non-Arabic
Train	50,715	10.42	13.34	6	13
Validation	2,199	0.36	9.60	0	1

Linguistic Analysis

Feature	Train Set	Validation Set
Top Words	في (20,250), و (16,977)	في (519), أنا (412)
Top Bigrams	(إن, أنا) (1,305)	(شاء, الله) (63)
Vocab Size	38,451	7,892
Unique Speakers	1,142	98

Train Distribution Validation Distribution
Word Count Distributions (Left: Train, Right: Validation)

How to Use ? 🧑‍💻

Loading with Hugging Face

from datasets import load_dataset
import IPython.display as ipd

# Load dataset (streaming recommended for large datasets)
ds = load_dataset('NightPrince/MasriSpeech-Full', 
                 split='train',
                 streaming=True)

# Get first sample
sample = next(iter(ds))
print(f"Transcript: {sample['transcription']}")

# Play audio
ipd.Audio(sample['audio']['array'], 
          rate=sample['audio']['sampling_rate'])

Preprocessing the Dataset

from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
from datasets import load_dataset
import torch
model_name = "facebook/wav2vec2-base-960h"  # Spanish example
# or "facebook/wav2vec2-large-xlsr-53-en" for English
processor = Wav2Vec2Processor.from_pretrained(model_name)
model = Wav2Vec2ForCTC.from_pretrained(model_name)

def prepare_dataset(batch):
    audio = batch["audio"]
    
    # Extract audio array and sampling rate
    audio_array = audio["array"]
    sampling_rate = audio["sampling_rate"]
    
    # Process audio using feature extractor only
    inputs = processor.feature_extractor(
        audio_array, 
        sampling_rate=sampling_rate, 
        return_tensors="pt"
    )
    
    batch["input_values"] = inputs.input_values[0]
    
    # Process transcription using tokenizer only
    labels = processor.tokenizer(
        batch["transcription"], 
        return_tensors="pt"
    )
    
    batch["labels"] = labels["input_ids"][0]
    
    return batch

# Apply preprocessing to the entire dataset
print("Processing entire dataset...")
dataset = ds.map(prepare_dataset, remove_columns=["audio", "transcription"])

Fine-Tuning an ASR Model

from transformers import AutoModelForCTC, TrainingArguments, Trainer

# Load pre-trained model
model = AutoModelForCTC.from_pretrained("facebook/wav2vec2-base-960h")

# Define training arguments
training_args = TrainingArguments(
    output_dir="./results",
    evaluation_strategy="epoch",
    learning_rate=2e-5,
    per_device_train_batch_size=16,
    num_train_epochs=3,
    save_steps=10,
    save_total_limit=2,
    logging_dir="./logs",
    logging_steps=10,
)

# Initialize Trainer
trainer = Trainer(
    model=model,
    args=training_args,
    train_dataset=dataset,
    eval_dataset=dataset,
)

# Train the model
trainer.train()

Evaluating the Model

# Evaluate the model
eval_results = trainer.evaluate()
print("Evaluation Results:", eval_results)

Exporting the Model

# Save the fine-tuned model
model.save_pretrained("./fine_tuned_model")
processor.save_pretrained("./fine_tuned_model")

📜 Citation

If you use MasriSpeech-Full in your research or work, please cite it as follows:

@dataset{masrispeech_full,
  author       = {Yahya Muhammad Alnwsany},
  title        = {MasriSpeech-Full: Large-Scale Egyptian Arabic Speech Corpus},
  year         = {2025},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/collections/NightPrince/masrispeech-dataset-68594e59e46fd12c723f1544}
}

📜 Licensing

This dataset is released under the Apache 2.0 License. You are free to use, modify, and distribute the dataset, provided you comply with the terms of the license. For more details, see the LICENSE.

🙌 Acknowledgments

We would like to thank the following for their contributions and support:

Annotators: For their meticulous work in creating high-quality transcriptions.
Hugging Face: For providing tools and hosting the dataset.
Open-Source Community: For their continuous support and feedback.

💡 Use Cases

MasriSpeech-Full can be used in various applications, including:

Automatic Speech Recognition (ASR) for Egyptian Arabic.
Dialectal Arabic linguistic research.
Speech synthesis and voice cloning.
Training and benchmarking machine learning models for low-resource languages.

🤝 Contributing

We welcome contributions to improve MasriSpeech-Full. If you have suggestions, find issues, or want to add new features, please:

Fork the repository.
Create a new branch for your changes.
Submit a pull request with a detailed description of your changes.

For questions or feedback, feel free to contact the maintainer.

📝 Changelog

[1.0.0] - 2025-08-02

Initial release of MasriSpeech-Full.
Includes 52,914 audio samples with transcriptions.
Train/validation splits provided.
Dataset hosted on Hugging Face.