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

2026-04-01

AI Summary

This work addresses the challenge of detecting partially spoofed audio—where synthetic segments are spliced into genuine recordings—by proposing an unsupervised method that requires no training, annotations, or model modifications. The approach leverages frame-level embedding trajectories from frozen foundation speech models and quantifies representational continuity via first-order differences, treating temporal smoothness as a universal forensic cue to effectively identify abrupt discontinuities at splicing boundaries. This paradigm demonstrates strong cross-model generalization, achieving an equal error rate (EER) of 8.08% on the PartialSpoof dataset—comparable to supervised methods—and significantly outperforming supervised baselines on LLM-generated LlamaPartialSpoof with an EER of 24.12%.

Technology Category

• Natural Language Processing: Speech
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Diffusion Models for Vision

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

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.

Problem

• partial audio deepfake
• splice detection
• speech foundation models
• training-free detection
• audio forensics

Innovation

• training-free
• embedding trajectory
• speech foundation models
• partial audio deepfake
• audio forensics