# TRACE: Training-Free Partial Audio Deepfake Detection via Embedding Trajectory Analysis of Speech Foundation Models

arXiv: https://arxiv.org/abs/2604.01083 · HTML: https://mlsanity.com/paper/2604.01083/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Builds on 30 of its 48 references that are in the map.
- First posted on arXiv 1 Apr 2026.

## Builds on (most cited first, 30 of 30)

- [Robust Speech Recognition via Large-Scale Weak Supervision](https://mlsanity.com/paper/2212.04356/) (2022)
- [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://mlsanity.com/paper/2006.11477/) (2020)
- [HuBERT: Self-Supervised Speech Representation Learning by Masked Prediction of Hidden Units](https://mlsanity.com/paper/2106.07447/) (2021)
- [WavLM: Large-Scale Self-Supervised Pre-Training for Full Stack Speech Processing](https://mlsanity.com/paper/2110.13900/) (2021)
- [XTTS: a Massively Multilingual Zero-Shot Text-to-Speech Model](https://mlsanity.com/paper/2406.04904/) (2024)
- [End-to-end anti-spoofing with RawNet2](https://mlsanity.com/paper/2011.01108/) (2020)
- [L2 Regularization versus Batch and Weight Normalization](https://mlsanity.com/paper/1706.05350/) (2017)
- [ADD 2022: the First Audio Deep Synthesis Detection Challenge](https://mlsanity.com/paper/2202.08433/) (2022)
- [ADD 2023: the Second Audio Deepfake Detection Challenge](https://mlsanity.com/paper/2305.13774/) (2023)
- [The PartialSpoof Database and Countermeasures for the Detection of Short Fake Speech Segments Embedded in an Utterance](https://mlsanity.com/paper/2204.05177/) (2022)
- [Half-Truth: A Partially Fake Audio Detection Dataset](https://mlsanity.com/paper/2104.03617/) (2021)
- [An Initial Investigation for Detecting Partially Spoofed Audio](https://mlsanity.com/paper/2104.02518/) (2021)
- [A Survey on Speech Deepfake Detection](https://mlsanity.com/paper/2404.13914/) (2024)
- [Nes2Net: A Lightweight Nested Architecture for Foundation Model Driven Speech Anti-spoofing](https://mlsanity.com/paper/2504.05657/) (2025)
- [Multi-Task Learning in Utterance-Level and Segmental-Level Spoof Detection](https://mlsanity.com/paper/2107.14132/) (2021)
- [LlamaPartialSpoof: An LLM-Driven Fake Speech Dataset Simulating Disinformation Generation](https://mlsanity.com/paper/2409.14743/) (2024)
- [An Efficient Temporary Deepfake Location Approach Based Embeddings for Partially Spoofed Audio Detection](https://mlsanity.com/paper/2309.03036/) (2023)
- [Frame-to-Utterance Convergence: A Spectra-Temporal Approach for Unified Spoofing Detection](https://mlsanity.com/paper/2309.09837/) (2023)
- [Enhancing Partially Spoofed Audio Localization with Boundary-aware Attention Mechanism](https://mlsanity.com/paper/2407.21611/) (2024)
- [Transferable Adversarial Attacks on Audio Deepfake Detection](https://mlsanity.com/paper/2501.11902/) (2025)
- [PartialEdit: Identifying Partial Deepfakes in the Era of Neural Speech Editing](https://mlsanity.com/paper/2506.02958/) (2025)
- [Forensic deepfake audio detection using segmental speech features](https://mlsanity.com/paper/2505.13847/) (2025)
- [Adversarial Attacks on Audio Deepfake Detection: A Benchmark and Comparative Study](https://mlsanity.com/paper/2509.07132/) (2025)
- [TADA: Training-free Attribution and Out-of-Domain Detection of Audio Deepfakes](https://mlsanity.com/paper/2506.05802/) (2025)
- [Manipulated Regions Localization For Partially Deepfake Audio: A Survey](https://mlsanity.com/paper/2506.14396/) (2025)
- [Frame-level Temporal Difference Learning for Partial Deepfake Speech Detection](https://mlsanity.com/paper/2507.15101/) (2025)
- [CTC-TTS: LLM-based dual-streaming text-to-speech with CTC alignment](https://mlsanity.com/paper/2602.19574/) (2026)
- [Zero-Day Audio DeepFake Detection via Retrieval Augmentation and Profile Matching](https://mlsanity.com/paper/2509.21728/) (2025)
- [NE-PADD: Leveraging Named Entity Knowledge for Robust Partial Audio Deepfake Detection via Attention Aggregation](https://mlsanity.com/paper/2509.03829/) (2025)
- [Localizing Speech Deepfakes Beyond Transitions via Segment-Aware Learning](https://mlsanity.com/paper/2601.21925/) (2026)

## Built on by (most cited first, 5 of 5)

- [Dual-Branch Gated Fusion for Open-Set Audio Deepfake Source Tracing](https://mlsanity.com/paper/2606.10223/) (2026)
- [Do Transformations Reveal the Truth? Generative Residual Learning for Generalized AI-Generated Image Detection](https://mlsanity.com/paper/2607.08674/) (2026)
- [Uncertainty-Aware Deepfake Detection via Multi-View Structural Learning](https://mlsanity.com/paper/2607.28769/) (2026)
- [Robust Spoofed Speech Detection via Temporal Pyramid Modeling](https://mlsanity.com/paper/2606.16837/) (2026)
- [Trajectory Dynamics in Self-Supervised Learning Latent Space for Audio Deepfake Detection](https://mlsanity.com/paper/2608.13817/) (2026)

## Abstract (arXiv)

Partial audio deepfakes, where synthesized segments are spliced into genuine recordings, are particularly deceptive because most of the audio remains authentic. Existing detectors are supervised: they require frame-level annotations, overfit to specific synthesis pipelines, and must be retrained as new generative models emerge. We argue that this supervision is unnecessary. We hypothesize that speech foundation models implicitly encode a forensic signal: genuine speech forms smooth, slowly varying embedding trajectories, while splice boundaries introduce abrupt disruptions in frame-level transitions. Building on this, we propose TRACE (Training-free Representation-based Audio Countermeasure via Embedding dynamics), a training-free framework that detects partial audio deepfakes by analyzing the first-order dynamics of frozen speech foundation model representations without any training, labeled data, or architectural modification. We evaluate TRACE on four benchmarks that span two languages using six speech foundation models. In PartialSpoof, TRACE achieves 8.08% EER, competitive with fine-tuned supervised baselines. In LlamaPartialSpoof, the most challenging benchmark featuring LLM-driven commercial synthesis, TRACE surpasses a supervised baseline outright (24.12% vs. 24.49% EER) without any target-domain data. These results show that temporal dynamics in speech foundation models provide an effective, generalize signal for training-free audio forensics.