README.md

metadata

license: cc-by-4.0
task_categories:
  - text-to-speech
  - automatic-speech-recognition
language:
  - ar
tags:
  - speech
  - audio
  - tts
  - arabic
  - ljspeech
size_categories:
  - 10K<n<100K

arabic-speech-dataset

Dataset Description

Arabic speech dataset for TTS training with diacritized and dediacritized text variants

Dataset Structure

This dataset follows the LJSpeech format with 3 columns:

• Column 1: Audio file identifier
• Column 2: Original Arabic text with diacritics
• Column 3: Processed Arabic text (mix of diacritized and dediacritized)

Data Fields

• audio: Audio file (WAV format, 16kHz)
• filename: Audio file identifier
• original_text: Original Arabic transcription with full diacritics
• processed_text: Processed transcription (60% dediacritized, 40% original)

Dataset Statistics

• Total samples: 91,176
• Audio format: WAV, 16kHz, mono
• Language: Arabic
• Text format: 3-column LJSpeech format

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("MrEzzat/arabic-speech-dataset")

# Access audio and text
for sample in dataset["train"]:
    audio = sample["audio"]
    original_text = sample["original_text"]
    processed_text = sample["processed_text"]
    filename = sample["filename"]

License

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