SpeechAntiSpoofingBenchmarks/ArAD
ArAD (Arabic Audio Deepfake, test split) Benchmark-ready packaging of the test split of the Arabic Audio Deepfake (ArAD) dataset: binary anti-spoofing on Arabic (primarily Levantine dialect) speech. Overview The source dataset (RVC v2 voice-conversion deepfakes vs. genuine Arabic recordings) is published at DeepFake-Audio-Rangers/Arabic_Audio_Deepfake. This repo packages only the official test split (3913 clips) into the Arena's canonical schema. Clips shorter… See the full description on the dataset page: https://huggingface.co/datasets/SpeechAntiSpoofingBenchmarks/ArAD.
| Author / organization | SpeechAntiSpoofingBenchmarks |
|---|---|
| License (as declared) | odc-by |
| Downloads | 1.1K (as reported by Hugging Face) |
| Likes | 1 |
| Created | 8 June 2026 |
| Last updated at source | 27 June 2026 |
Popularity, last 3 days
1,132 now (+24 since 2026-10-16). Counts as reported by the source, snapshotted daily by Silytics.
More from SpeechAntiSpoofingBenchmarks
Silytics shows public metadata from Hugging Face and links to the original. Information may be incomplete or out of date. Last refreshed 18 October 2026. Licenses and terms of the original apply to the work itself. Not financial, legal or professional advice.