Search papers, labs, and topics across Lattice.
This paper addresses the challenge of detecting partial deepfake speech, which undermines existing detection systems as the manipulated segments become less prominent. By leveraging a self-embedding steganography technique, the authors create a proactive defense that allows for the extraction of reference content from clean speech signals, enhancing detection capabilities. Experimental results indicate that this training-free method effectively complements traditional passive defenses, offering a robust solution to the growing issue of audio manipulation.
A training-free self-embedding strategy can significantly enhance the detection of partial deepfake speech, even when only small segments are manipulated.
Partial deepfake speech, where only limited segments of an utterance are synthesized or manipulated, poses a significant challenge to existing deepfake detection systems. As the proportion of spoofed regions decreases, passive detectors become increasingly unreliable, and accurate detection and restoration remain challenging. In this paper, we revisit audio steganography from a new perspective and propose its use as a proactive defense against partially deepfaked audio. In particular, we consider a self-embedding strategy in which a clean speech signal embeds a compressed representation of itself, enabling post-hoc extraction of reference content. We demonstrate how existing audio steganography methods can be repurposed to support detection of partial deepfakes through codec-based restoration. Experiments on a benchmark dataset show that the proposed approach complements passive defenses. Remarkably, the proposed method operates without any training, providing a robust and data-efficient alternative for partial deepfake detection.
