Deepfake Detection

The Phonexia Deepfake Detection technology is designed to detect speaker verification spoofing and deepfake audio using the wav2vec 2.0 XLSR model. This approach enhances the reliability of speaker verification systems against basic spoofing attacks and deepfake audio. The model is primarily trained on ASVspoof datasets, which include a wide range of synthesized, converted, and replayed speech examples.

The technology leverages self-supervised learning and data augmentation techniques, this helps to achieve adequate performance in detecting fraudulent audio inputs, thereby strengthening the security of voice-based authentication systems. The model requires a minimum of 4 seconds of speech for inference.