# Alethia: A Foundational Encoder for Voice Deepfakes Source: [Alethia: A Foundational Encoder for Voice Deepfakes]() Collection index: [llms.txt]() Data updated: 2026-05-05T05:00:04.460Z Authors: Yi Zhu, Brahmi Dwivedi, Jayaram Raghuram, Surya Koppisetti Published: 2026-04-30T21:32:40.000Z ## 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 $5$ different tasks with $56$ 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. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Speech Deepfake Detection on SDD-Eval-50 All](): Speech Deepfake Detection; SDD-Eval-50 All; EER: 5.2. - [Speech Deepfake Detection on SDD-Eval-50 Challenging](): Speech Deepfake Detection; SDD-Eval-50 Challenging; EER: 11.5. - [Audio-Visual Deepfake Detection on PolyGlotFake](): Audio-Visual Deepfake Detection; PolyGlotFake; EER: 7.1. - [Audio-Visual Deepfake Detection on PolyGlotFake (zero-shot)](): Audio-Visual Deepfake Detection; PolyGlotFake zero-shot; EER (zero-shot): 7.1. - [Source Tracing on ASVspoof5-ST (test)](): Source Tracing; ASVspoof5-ST (test); Silhouette Score: 0.02. - [Partially Fake Speech Localization on Half-Truth (HT)](): Partially Fake Speech Localization; Half-Truth (HT); EER: 9.2. - [Partially Fake Speech Localization on LlamaPartialSpoof (LPS)](): Partially Fake Speech Localization; LlamaPartialSpoof (LPS); EER: 19.8. - [Audio-Visual Deepfake Detection on FakeAVCeleb (EER)](): Audio-Visual Deepfake Detection; FakeAVCeleb; EER: 6.3. - [Audio-Visual Deepfake Detection on FakeAVCeleb zero-shot](): Audio-Visual Deepfake Detection; FakeAVCeleb zero-shot; EER (zero-shot): 6.3. - [Partially Fake Speech Localization on PartialSpoof (PS)](): Partially Fake Speech Localization; PartialSpoof (PS); EER: 27.1. - [Singing Voice Deepfake Detection on CtrSVDD](): Singing Voice Deepfake Detection; CtrSVDD; EER: 10.8. ## Original sources and implementation [Original paper]() [PDF]()