Dataset · Mohammed01
Benchmark for Arabic speech deepfake detection
Mohammed01/ArFake
First multi-dialect dataset of synthetic and real Arabic speech for detecting audio fakes.
What's inside
It contains Arabic audio recordings and texts spanning multiple dialects. The collection volume ranges from 10,000 to 100,000 records, including specially generated synthetic speech. The data is assembled as the first comprehensive benchmark to test model robustness on unfamiliar generators and dialects.
What you can do with it
- Training spoofing and voice deepfake detectors in Arabic
- Testing audio classification systems for robustness against unfamiliar dialects and speech synthesizers
- Testing ASR and speech synthesis models on Arabic dialects
- Tasks
- Audio classification, Automatic speech recognition, Text → audio
- Languages
- Arabic
- Size
- 10K–100K rows
- Downloads this month
- 10.2K
- Likes
- 9
- Access
- on request on Hugging Face
How to load it
from datasets import load_dataset
ds = load_dataset("Mohammed01/ArFake")