A curated Arabic speech dataset combining three sources, transcribed via Whisper-large-v3 (primary) and omniASR-7B (fallback). Built for Arabic TTS and ASR training.

Quick Start

from datasets import load_dataset

ds = load_dataset("KFUPM-JRCAI/arabic_speech", split="train") print(ds[0]["transcription"]) # ASR text print(ds[0]["audio"]["array"]) # audio waveform (float32, 16kHz) print(ds[0]["timestamps"]) # word-level… See the full description on the dataset page: https://huggingface.co/datasets/KFUPM-JRCAI/arabic_speech.