Arabic-STT/kejue-voice

Kejue Voice

Unified Arabic speech dataset with transcriptions and continuations, organized by dialect.

Overview

• 1.6M+ samples across 18 source datasets
• 17 dialect configs (ar, ar-eg, ar-sa, ar-lb, ar-bh, ar-iq, ar-ae, ar-jo, ar-kw, ar-ma, ar-om, ar-ps, ar-qa, ar-sy, ar-tn, ar-dz, ar-en)
• Standardized schema: id, audio, text, continuation, dialect, source
• Audio + text transcription + generated continuation for each sample

Usage

from datasets import load_dataset

# Load all Arabic (general/MSA)
ds = load_dataset("KejueAI/kejue-voice", "ar")

# Load specific dialect
ds_eg = load_dataset("KejueAI/kejue-voice", "ar-eg")  # Egyptian
ds_sa = load_dataset("KejueAI/kejue-voice", "ar-sa")  # Saudi
ds_lb = load_dataset("KejueAI/kejue-voice", "ar-lb")  # Lebanese

# Filter by source dataset
adi17_eg = ds_eg.filter(lambda x: x["source"] == "adi17")

Dialect Configs

Config	Dialect	Rows
ar	Arabic (general/MSA)	~628K
ar-eg	Egyptian	~172K
ar-iq	Iraqi	~277K
ar-sa	Saudi	~159K
ar-bh	Bahraini/Gulf	~90K
ar-jo	Jordanian	~41K
ar-sy	Syrian/Levantine	~37K
ar-ps	Palestinian	~35K
ar-lb	Lebanese	~35K
ar-ae	Emirati	~44K
ar-kw	Kuwaiti	~28K
ar-qa	Qatari	~24K
ar-ma	Moroccan	~18K
ar-om	Omani	~17K
ar-en	Arabic-English code-switching	~12K
ar-tn	Tunisian	~3K
ar-dz	Algerian	~1K

Source Datasets

Source	Original Dataset	Rows
adi17	KejueAI/kejue-voice-adi17	~650K
mgb2	MohamedRashad/mgb2-arabic	~374K
sada22	MohamedRashad/SADA22	~232K
arabic-youtube	KejueAI/arabic-youtube-corpus-with-continuations	~226K
bahraini-speech	KejueAI/bahraini-speech-continuations-v3	~87K
arabic-voices-clean	KejueAI/arabic-voices-clean-v4-continuations	~23K
code-switching	MohamedRashad/arabic-english-code-switching	~12K
omnilingual-*	facebook/omnilingual-asr-corpus	~15K
mgb3	MohamedRashad/MGB-3-Arabic	~2K
sawtarabi	ArabicSpeech/sawtarabi	~2.5K
lebanese-youtube	KejueAI/lebanese-youtube-corpus	~1.7K

Schema

Column	Type	Description
id	string	Unique row identifier
audio	Audio	Audio sample
text	string	Transcription
continuation	string	Generated continuation (may be empty)
dialect	string	Dialect code (ar-eg, ar-sa, etc.)
source	string	Original dataset name

Post-processing

All data has been cleaned:

• Rows with repeated words (3x consecutive) removed from text and continuation
• Overlapping words between end of text and start of continuation stripped

Removed: overlap with the Open Universal Arabic ASR Leaderboard test sets

On 2026-10-14 we removed 4,904 rows (21.8 h) from 510 of the 1163 parquet files. Each removed row overlaps the test material of the Open Universal Arabic ASR Leaderboard (test sets with hits: casablanca, casablanca_validation, common_voice, masc_clean, masc_noisy, mgb2, sada).

• How rows were matched: source ids, normalised text, and audio fingerprints on 100% of the rows (contamination guard version 2, arabic-english-stt #8).
• Row ids: the removed rows are listed in contamination/removed_ids.txt.
• Counts per test set: in contamination/summary.json.
• Unchanged files: all other files are unchanged.
• Earlier versions: earlier revisions of this repo keep the original files.