fr3on/egyptian-songs

Egyptian Arabic Songs Dataset 🎵

Dataset Description

This dataset contains 3,063 lines of Egyptian Arabic song lyrics with English translations, spanning 280 songs from 1983-2021. The dataset features natural Egyptian Arabic dialect (العامية المصرية) as used in popular music, with automatic genre classification and line-type detection.

Languages

• Source: Egyptian Arabic (ar_EG) - Colloquial dialect in music
• Target: English (en)

Dataset Summary

Egyptian Arabic music represents a rich cultural tradition and is one of the most influential Arabic music styles globally. This dataset provides:

• Authentic Egyptian Arabic lyrics from popular songs
• Poetic and colloquial expressions
• Genre-classified content for theme-aware training
• Temporal context (release years)
• Song and album metadata for context

Dataset Structure

Data Format

Each entry contains:

{
  "id": "song0001_line0042",
  "arabic": "المدينة قلبها زحمة بالهموم",
  "english": "The heart of the city teems with sorrow.",
  "song_name": "المدينة",
  "album_name": "يا طريق",
  "year": 1983,
  "song_id": 1,
  "dialect": "egyptian",
  "language": "ar",
  "language_variant": "ar_EG",
  "content_type": "song_lyrics",
  "line_type": "verse",
  "genre": "sadness"
}

Data Fields

Field	Type	Description
id	string	Unique identifier (format: songXXXX_lineYYYY)
arabic	string	Egyptian Arabic lyrics line
english	string	English translation
song_name	string	Song title (Arabic)
album_name	string	Album name (Arabic)
year	int	Release year
song_id	int	Unique song identifier
dialect	string	Dialect identifier (always "egyptian")
language	string	ISO language code (always "ar")
language_variant	string	Specific variant code (always "ar_EG")
content_type	string	Content type (always "song_lyrics")
line_type	string	Line classification (verse/chorus/instrumental)
genre	string	Auto-detected music theme/genre

Dataset Statistics

Overview

• Total Lines: 3,063
• Unique Songs: 280
• Unique Albums: 35
• Year Range: 1983-2021
• Unique Genres: 9
• Average Arabic Length: 28.6 characters
• Average English Length: 49.2 characters

Genre Distribution

Genre	Count	Percentage
general	1,288	42.1%
romance	1,090	35.6%
social	224	7.3%
hope	128	4.2%
nostalgia	110	3.6%
sadness	77	2.5%
celebration	71	2.3%
spiritual	59	1.9%
patriotic	16	0.5%

Line Type Distribution

• verse: 2,458 (80.2%)
• chorus: 605 (19.8%)

Temporal Distribution

Year	Lines
1983	115
1984	82
1986	99
1987	72
1988	49
1989	78
1990	55
1991	66
1992	98
1993	81
1994	84
1995	49
1996	73
1998	71
1999	58
2000	80
2001	120
2003	112
2004	128
2005	108
2007	111
2009	130
2010	17
2011	129
2012	10
2013	125
2014	118
2016	122
2017	107
2018	156
2020	218
2021	142

Top Songs by Line Count

Song Name	Lines
فاكرني يا حب	29
يوم تلات	27
عم الطبيب	22
الزمن	21
أيام وبنعيشها	21
هدد	19
أول كل حاجة	19
إلا هي	18
قالتلي قول	18
وأنا معاك	17

Genres Explained

This dataset includes automatic genre/theme classification using keyword-based detection:

• romance - Love songs, relationships, affection
• sadness - Melancholic themes, heartbreak, sorrow
• nostalgia - Memories, past times, longing
• hope - Optimistic themes, dreams, aspirations
• social - Social commentary, life observations
• patriotic - National pride, homeland themes
• spiritual - Religious or spiritual content
• celebration - Joyful occasions, festivities
• general - General themes without specific classification

Use Cases

✅ Recommended Use Cases

• Egyptian Arabic Translation: Train translation models for colloquial Egyptian
• Lyric Generation: Generate Egyptian Arabic song lyrics
• Genre-Aware Models: Train models that understand musical themes
• Cultural NLP: Study Egyptian culture through music
• Dialect Research: Research Egyptian Arabic characteristics in artistic contexts
• Music Information Retrieval: Lyrics-based music analysis
• Cross-lingual Understanding: Arabic-English semantic alignment

⚠️ Limitations

• Domain Scope: Limited to song lyrics (poetic/artistic language)
• Size: 3,063 lines (moderate size for specialized tasks)
• Time Period: 1983-2021 (may not reflect modern slang)
• Artist Coverage: Limited to specific artists/albums
• Line Context: Individual lines may lack full song context
• Genre Accuracy: Automatic genre detection may have errors

Loading the Dataset

Using Hugging Face Datasets

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("fr3on/egyptian-songs")

# Access the data
print(dataset['train'][0])

# Filter by genre
romance_songs = dataset['train'].filter(lambda x: x['genre'] == 'romance')

# Filter by year
songs_80s = dataset['train'].filter(lambda x: 1980 <= x['year'] < 1990)

# Filter by song
specific_song = dataset['train'].filter(lambda x: x['song_id'] == 1)

Using Pandas

import pandas as pd

# Load Parquet file directly
df = pd.read_parquet("data/train-00000-of-00001.parquet")

# Analyze genres
print(df['genre'].value_counts())

# Get all lines from a specific song
song_lines = df[df['song_name'] == 'المدينة'].sort_values('id')

# Filter by year range
df_90s = df[(df['year'] >= 1990) & (df['year'] < 2000)]

Training Examples

Translation Model (Lyric-Aware)

from datasets import load_dataset
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, Seq2SeqTrainer

# Load dataset
dataset = load_dataset("fr3on/egyptian-songs")

# Load model for Arabic-English translation
model_name = "Helsinki-NLP/opus-mt-ar-en"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForSeq2SeqLM.from_pretrained(model_name)

# Tokenize
def preprocess(examples):
    inputs = tokenizer(examples['arabic'], truncation=True, max_length=128)
    targets = tokenizer(examples['english'], truncation=True, max_length=128)
    inputs['labels'] = targets['input_ids']
    return inputs

tokenized = dataset.map(preprocess, batched=True)

# Train
trainer = Seq2SeqTrainer(
    model=model,
    train_dataset=tokenized['train'],
    eval_dataset=tokenized['test']
)
trainer.train()

Genre-Aware Text Generation

from datasets import load_dataset
from transformers import GPT2LMHeadModel, GPT2Tokenizer

dataset = load_dataset("fr3on/egyptian-songs")

# Train separate models per genre
for genre in ['romance', 'sadness', 'hope']:
    genre_data = dataset['train'].filter(lambda x: x['genre'] == genre)
    print(f"Training {genre} model with {len(genre_data)} examples")
    # Train genre-specific model

Lyric Generation

from datasets import load_dataset

dataset = load_dataset("fr3on/egyptian-songs")

# Prepare for language modeling
def prepare_for_lm(example):
    return {'text': f"<|song|>{example['song_name']}<|genre|>{example['genre']}<|lyrics|>{example['arabic']}<|endoftext|>"}

lm_dataset = dataset.map(prepare_for_lm)
# Train GPT-style model for lyric generation

Data Collection & Processing

Source

• Origin: Egyptian song lyrics with professional translations
• Quality: Human-curated translations
• Cultural Context: Authentic Egyptian musical heritage

Processing Pipeline

1. Extraction: Load from Excel with metadata
2. Cleaning: Remove empty/short lines, whitespace normalization
3. Deduplication: Hash-based duplicate removal
4. Genre Detection: Automatic classification using keyword matching
5. Line Type Detection: Classify verses, chorus, instrumental sections
6. Validation: Quality checks and statistics generation
7. Format Conversion: Export to JSONL and Parquet

Data Quality

• ✅ Deduplicated using MD5 hash matching
• ✅ Filtered lines < 3 characters
• ✅ Removed rows with missing translations
• ✅ Normalized whitespace
• ✅ Validated Arabic-English text pairs
• ✅ Preserved song/album context

Considerations for Using the Data

Egyptian Arabic in Music

Egyptian Arabic in songs has unique characteristics:

• Poetic Language: More metaphorical and artistic than everyday speech
• Rhyme Schemes: Traditional Arabic poetry patterns (قافية)
• Colloquialisms: Mix of formal and informal expressions
• Cultural References: Egyptian-specific idioms and references
• Emotional Intensity: Heightened expressive language

Recommended Training Approaches

1. Fine-tune on this dataset after pre-training on general Egyptian Arabic
2. Combine with speech data for richer dialect representation
3. Use genre filtering for domain-specific applications
4. Preserve song context when possible for narrative understanding
5. Consider temporal aspects - language evolves over decades
6. Balance genres if training general-purpose models

Ethical Considerations

• Copyright: Original songs are copyrighted; dataset for research purposes
• Cultural Sensitivity: Respect Egyptian cultural and musical heritage
• Attribution: Credit original artists and composers
• Representation: One slice of Egyptian musical culture
• Usage: For educational, research, and non-commercial purposes

License

This dataset is released under the CC-BY-4.0 License.

⚠️ Note: Original songs are copyrighted. This dataset is intended for research, education, and non-commercial use only.

Citation

If you use this dataset in your research, please cite:

@dataset{egyptian_songs_2025,
  title={Egyptian Arabic Songs Dataset},
  author={fr3on},
  year={2025},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/fr3on/egyptian-songs}
}

Acknowledgments

• Source: Egyptian song lyrics with translations
• Processing: Automatic genre and line-type detection
• Cultural Heritage: Egyptian musical tradition

Version History

• v1.0.0 (2025-12-18): Initial release
  • 3,063 lyric lines
  • 280 songs from 35 albums
  • 1983-2021 temporal coverage
  • 9 genre categories
  • Automatic genre and line-type detection

---

Keywords: Egyptian Arabic, ar_EG, dialect, colloquial, music, lyrics, songs, translation, NLP, Egyptian culture, Arabic poetry, song generation

Dataset Size: 3,063 lines | Format: Parquet | License: CC-BY-4.0