What Do Deepfake Speech Detectors Actually Hear?

• URL: http://arxiv.org/abs/2606.10912v1
• Date: Tue, 09 Jun 2026 14:21:45 GMT
• Title: What Do Deepfake Speech Detectors Actually Hear?
• Abstract summary: Deepfake speech detectors often output a single score without explaining why an audio sample is flagged, where in the signal the evidence lies, or what cues drive the decision.<n>We propose an audio-native explainability pipeline using Integrated Gradients on time-aligned self-supervised representations to localize decision evidence over time.<n>We apply the proposed method to three WavLM-based detectors (AASIST, CA-MHFA, SLS) on ASVspoof 5 and manually annotate the highest-attribution regions to provide a semantic meaning of the most important cues.
• Score: 0.16419687521433918
• License: http://creativecommons.org/licenses/by-sa/4.0/
• Abstract: Deepfake speech detectors often output a single score without explaining why an audio sample is flagged, where in the signal the evidence lies, or what cues drive the decision. We propose an audio-native explainability pipeline using Integrated Gradients on time-aligned self-supervised representations to localize decision evidence over time. We apply the proposed method to three WavLM-based detectors (AASIST, CA-MHFA, SLS) on ASVspoof 5 and manually annotate the highest-attribution regions to provide a semantic meaning of the most important cues. Despite similar performance, the detectors rely on different cues: AASIST emphasizes non-speech/environment cues, CA-MHFA focuses on localized phoneme artifacts, and SLS relies on word boundaries and spectral integrity. We move beyond speculative reasoning and validate our findings by causal masking of the primary detector cues. Observed performance degradation further supports the explained detector semantics.

Related papers

• An Intervention-Based Framework for Shortcut Diagnosis in Spoofing Countermeasures [5.593982292458519]
  We propose an intervention-based diagnostic framework to identify confound-driven shortcut dependencies.<n>We operationalise this through controlled acoustic perturbations targeting non-speech structure, spectral content, and signal energy.<n>Results reveal that non-speech interventions produce the largest performance shifts, confirming non-speech intervals as a dominant shortcut.
  arXiv  Detail & Related papers  (2026-07-03T09:42:31Z)
• Detecting Hallucinations in SpeechLLMs at Inference Time Using Attention Maps [1.8825446478246406]
  Hallucinations in Speech Large Language Models (SpeechLLMs) pose significant risks, yet existing detection methods rely on gold-standard outputs that are costly or impractical to obtain.<n>We investigate four attention-derived metrics: AUDIORATIO, AUDIOCONSISTENCY, AUDIOENTROPY, and TEXTENTROPY, designed to capture pathological attention patterns associated with hallucination.<n>Across automatic speech recognition and speech-to-text translation tasks, our approach outperforms uncertainty-based and prior attention-based baselines.
  arXiv  Detail & Related papers  (2026-04-21T15:18:10Z)
• ICLAD: In-Context Learning with Comparison-Guidance for Audio Deepfake Detection [7.887913982488066]
  We introduce a novel textbfIn-textbfContext textbfLearning paradigm with comparison-guidance for textbfAudio textbfDeepfake detection.<n>The framework enables the use of audio language models (ALMs) for training-free generalization to unseen deepfakes.
  arXiv  Detail & Related papers  (2026-04-17T23:44:33Z)
• SNAP: Speaker Nulling for Artifact Projection in Speech Deepfake Detection [51.096014381455454]
  Self-supervised learning-based speech encoders struggle to generalize across unseen speakers.<n>Speaker entanglement causes detectors to exploit speaker-specific correlations rather than artifact-related cues.<n>We introduce SNAP, a speaker-nulling framework, to mitigate this reliance.
  arXiv  Detail & Related papers  (2026-03-21T07:05:30Z)
• Causal Tracing of Audio-Text Fusion in Large Audio Language Models [48.849764961130795]
  We adapt causal tracing to investigate the internal information flow of LALMs during audio comprehension.<n>By conducting layer-wise and token-wise analyses across DeSTA, Qwen, and Voxtral, we evaluate the causal effects of individual hidden states.
  arXiv  Detail & Related papers  (2026-03-14T05:40:59Z)
• Towards Explicit Acoustic Evidence Perception in Audio LLMs for Speech Deepfake Detection [23.695892348165497]
  Speech deepfake detection (SDD) focuses on identifying whether a given speech signal is genuine or has been synthetically generated.<n>Existing audio large language model (LLM)-based methods are often biased toward semantically correlated cues.<n>We introduce SDD with Auditory Perception-enhanced Audio Large Language Model (SDD-APALLM)
  arXiv  Detail & Related papers  (2026-01-30T15:16:43Z)
• Toward Faithful Retrieval-Augmented Generation with Sparse Autoencoders [39.5490415037017]
  Retrieval-Augmented Generation (RAG) improves the factuality of large language models (LLMs) by grounding outputs in retrieved evidence.<n>Existing hallucination detection methods for RAG often rely on large-scale detector training.<n>We introduce RAGLens, a lightweight hallucination detector that accurately flags unfaithful RAG outputs.
  arXiv  Detail & Related papers  (2025-12-09T18:33:22Z)
• Sparse deepfake detection promotes better disentanglement [4.901409400999413]
  We show that sparse deepfake detection can improve detection performance, with an EER of 23.36% on ASVSpoof5 test set, with 95% of sparsity.<n>We then show that these representations provide better disentanglement, using completeness and modularity metrics based on mutual information.
  arXiv  Detail & Related papers  (2025-10-07T09:03:39Z)
• Generalizable Speech Deepfake Detection via Information Bottleneck Enhanced Adversarial Alignment [48.73836179661632]
  Confidence-guided adversarial alignment adaptively suppresses attack-specific artifacts without erasing discriminative cues.<n>IB-CAAN consistently outperforms baseline and state-of-the-art performance on many benchmarks.
  arXiv  Detail & Related papers  (2025-09-28T03:48:49Z)
• Coarse-to-Fine Proposal Refinement Framework for Audio Temporal Forgery Detection and Localization [60.899082019130766]
  We introduce a frame-level detection network (FDN) and a proposal refinement network (PRN) for audio temporal forgery detection and localization. FDN aims to mine informative inconsistency cues between real and fake frames to obtain discriminative features that are beneficial for roughly indicating forgery regions. PRN is responsible for predicting confidence scores and regression offsets to refine the coarse-grained proposals derived from the FDN.
  arXiv  Detail & Related papers  (2024-07-23T15:07:52Z)
• Introducing Semantics into Speech Encoders [91.37001512418111]
  We propose an unsupervised way of incorporating semantic information from large language models into self-supervised speech encoders without labeled audio transcriptions. Our approach achieves similar performance as supervised methods trained on over 100 hours of labeled audio transcripts.
  arXiv  Detail & Related papers  (2022-11-15T18:44:28Z)
• Spotting adversarial samples for speaker verification by neural vocoders [102.1486475058963]
  We adopt neural vocoders to spot adversarial samples for automatic speaker verification (ASV) We find that the difference between the ASV scores for the original and re-synthesize audio is a good indicator for discrimination between genuine and adversarial samples. Our codes will be made open-source for future works to do comparison.
  arXiv  Detail & Related papers  (2021-07-01T08:58:16Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.