English

Alethia: A Foundational Encoder for Voice Deepfakes

Sound 2026-05-04 v1 Computation and Language Audio and Speech Processing

Abstract

Existing voice deepfake detection and localization models rely heavily on representations extracted from speech foundation models (SFMs). However, downstream finetuning has now reached a state of diminishing returns. In this paper, we shift the focus to pretraining and propose a novel recipe that combines bottleneck masked embedding prediction with flow-matching based spectrogram reconstruction. The outcome, Alethia, is the first foundational audio encoder for various voice deepfake detection and localization tasks. We evaluate on 55 different tasks with 5656 benchmark datasets, and note Alethia significantly outperforms state-of-the-art SFMs with superior robustness to real-world perturbations and zero-shot generalization to unseen domains (e.g., singing deepfakes). We also demonstrate the limitation of discrete targets in masked token prediction, and show the importance of continuous embedding prediction and generative pretraining for capturing deepfake artifacts.

Cite

@article{arxiv.2605.00251,
  title  = {Alethia: A Foundational Encoder for Voice Deepfakes},
  author = {Yi Zhu and Brahmi Dwivedi and Jayaram Raghuram and Surya Koppisetti},
  journal= {arXiv preprint arXiv:2605.00251},
  year   = {2026}
}

Comments

Accepted to ICML 2026

R2 v1 2026-07-01T12:44:33.695Z