# Trajectory Dynamics in Self-Supervised Learning Latent Space for Audio Deepfake Detection

arXiv: https://arxiv.org/abs/2608.13817 · HTML: https://mlsanity.com/paper/2608.13817/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 19 of its 24 references that are in the map.
- First posted on arXiv 13 Aug 2026.

## Builds on (most cited first, 19 of 19)

- [wav2vec 2.0: A Framework for Self-Supervised Learning of Speech Representations](https://mlsanity.com/paper/2006.11477/) (2020)
- [XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale](https://mlsanity.com/paper/2111.09296/) (2021)
- [Does Audio Deepfake Detection Generalize?](https://mlsanity.com/paper/2203.16263/) (2022)
- [ASVspoof 2021: Towards Spoofed and Deepfake Speech Detection in the Wild](https://mlsanity.com/paper/2210.02437/) (2022)
- [One-class Learning Towards Synthetic Voice Spoofing Detection](https://mlsanity.com/paper/2010.13995/) (2020)
- [MLAAD: The Multi-Language Audio Anti-Spoofing Dataset](https://mlsanity.com/paper/2401.09512/) (2024)
- [Deepfake-Eval-2024: A Multi-Modal In-the-Wild Benchmark of Deepfakes Circulated in 2024](https://mlsanity.com/paper/2503.02857/) (2025)
- [Speech is Silver, Silence is Golden: What do ASVspoof-trained Models Really Learn?](https://mlsanity.com/paper/2106.12914/) (2021)
- [The Codecfake Dataset and Countermeasures for the Universally Detection of Deepfake Audio](https://mlsanity.com/paper/2405.04880/) (2024)
- [SAMO: Speaker Attractor Multi-Center One-Class Learning for Voice Anti-Spoofing](https://mlsanity.com/paper/2211.02718/) (2022)
- [Post-training for Deepfake Speech Detection](https://mlsanity.com/paper/2506.21090/) (2025)
- [SLIM: Style-Linguistics Mismatch Model for Generalized Audio Deepfake Detection](https://mlsanity.com/paper/2407.18517/) (2024)
- [Generalizable Audio Deepfake Detection via Latent Space Refinement and Augmentation](https://mlsanity.com/paper/2501.14240/) (2025)
- [TRACE: Training-Free Partial Audio Deepfake Detection via Embedding Trajectory Analysis of Speech Foundation Models](https://mlsanity.com/paper/2604.01083/) (2026)
- [Improving Out-of-Domain Audio Deepfake Detection via Layer Selection and Fusion of SSL-Based Countermeasures](https://mlsanity.com/paper/2509.12003/) (2025)
- [BreathNet: Generalizable Audio Deepfake Detection via Breath-Cue-Guided Feature Refinement](https://mlsanity.com/paper/2602.13596/) (2026)
- [ProSDD: Learning Prosodic Representations for Speech Deepfake Detection against Expressive and Emotional Attacks](https://mlsanity.com/paper/2604.13229/) (2026)
- [QAMO: Quality-aware Multi-centroid One-class Learning For Speech Deepfake Detection](https://mlsanity.com/paper/2509.20679/) (2025)
- [Fine-Grained Frame Modeling in Multi-head Self-Attention for Speech Deepfake Detection](https://mlsanity.com/paper/2602.04702/) (2026)

## Abstract (arXiv)

Human speech production is constrained by physiology, giving rise to characteristic temporal structure on acoustic signals. We hypothesise that these constraints manifest as structured trajectory dynamics in the latent space of Self-Supervised Learning (SSL) models, and that synthetic speech violates them detectably. To test this hypothesis, we train a causal Long Short-Term Memory (LSTM) next-frame predictor on bonafide speech only (Stage 1), using the deepfake-specialised SSL backbone Wav2Vec2-Large-AntiDeepfake, and compare against a static global-average-pooling baseline using identical features, thus isolating the contribution of temporal modelling. A supervised Stage 2, which trains a Multi-Layer Perceptron on the frozen LSTM internal states using labelled data, is included to characterise the role of spoof supervision. Our system achieves competitive or state-of-the-art performance across six benchmarks: ASVspoof 2019/2021, Codecfake, In-the-Wild, MLAAD-EN, and Deepfake-Eval-2024, including best published EER on ASVspoof 2021 (0.75%) and, notably, Stage 1 trained on bonafide speech only surpasses the published supervised baseline from the same backbone on DE2024 (30.35%). On near-domain benchmarks, static and dynamic approaches perform comparably. On harder cross-corpus benchmarks with diverse synthesis methods, trajectory dynamics provide substantial gains, confirming that temporal physiological constraints carry detection signal beyond utterance-level statistics.