Dataset · garystafford
Balanced audio dataset for deepfake detection
garystafford/deepfake-audio-detection
1,866 English audio recordings (evenly split between real and synthesized) to train voice deepfake detectors.
What's inside
The version 4 dataset consists of 1,866 audio samples in English: exactly 933 live human speech recordings and 933 synthesized audio examples. Synthetic voices were generated by six different TTS systems (including Hume AI added in this version).
What you can do with it
- Training binary classifiers to recognize synthesized speech (audio deepfakes)
- Testing anti-spoofing systems for robustness against modern TTS generators
- Tasks
- Audio classification
- Languages
- English
- Size
- 1K–10K rows
- Downloads this month
- 1.4K
- Likes
- 10
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("garystafford/deepfake-audio-detection")