license: apache-2.0
Quran TTS Dataset
Dataset Description
This is a comprehensive Text-to-Speech (TTS) dataset for the Arabic language,specifically tailored to the verses of the Holy Quran. The audio files have been generated using Premium Google text-to-speech engines and aligned with their corresponding Uthmani script transcriptions. The dataset contains 19 voices of male and female reciters with 114 surah
Please note that verses like الم and كهيعص and other verses alhuruf motaqati3a are bad generateds. You can locate each verse with: {reciter_id}_{surah}_1
Dataset Structure:
The dataset is composed of multiple Parquet files (.parquet), each containing a slice of the total data. Each row in the dataset represents a single audio clip and includes the following columns:
audio: The audio data, including the raw bytes and sampling rate (wav 16kHz).
duration: The duration of the audio clip in seconds (float).
text: The Arabic transcription of the verse in Uthmani script (string).
reciter: The name of the synthetic voice used for the recitation (string).
audio_id: A unique identifier for the audio file, typically in the format {reciter_id}{surah}{ayah} (string).
Audio Durations Analysis:
Based on an analysis of the duration column, here are some key statistics about the dataset:
General Statistics
Total Audio Files Analyzed: 131,314
Total Cumulative Duration: 223h 08m 37s
Minimum Duration: 0.26 seconds
Maximum Duration: 30.35 seconds
Average Duration: 6.12 seconds
Distribution by Duration Range
| Duration | File Count | Percentage |
|---|---|---|
| Less than 5s | 53,672 | 40.87% |
| 5s to 10s | 62,490 | 47.59% |
| 10s to 15s | 13,321 | 10.14% |
| 15s to 20s | 1,525 | 1.16% |
| 20s to 30s | 303 | 0.23% |
| More than 30s | 3 | 0.00% |