SONAR: Spectral-Contrastive Audio Residuals for Generalizable Deepfake Detection

Ido Nitzan Hidekel, Gal Lifshitz, Khen Cohen, Dan Raviv
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:42998-43012, 2026.

Abstract

Deepfake audio detectors often fail to generalize to unseen attacks, in part due to spectral bias: neural networks prioritize low-frequency structure while under-exploiting subtle high-frequency (HF) artifacts left by generative models. We introduce SONAR (Spectral-cONtrastive Audio Residuals), a frequency-guided framework that explicitly enforces representation-level consistency between semantic content and HF residuals. Unlike prior frequency-aware or dual-stream detectors that treat HF cues as auxiliary features, SONAR encourages structured interaction between content and noise representations in latent space. The model employs a dual-path architecture in which an XLSR encoder captures low-frequency content, while a parallel branch with learnable, value-constrained 1D SRM (Spatial Rich Model) high-pass filters distills HF residuals. The two representations are fused via frequency cross-attention and trained with a Jensen–Shannon alignment loss that promotes LF–HF consistency for genuine audio and amplifies inconsistency for deepfakes. Evaluated on ASVspoof 2021 and in-the-wild benchmarks, SONAR achieves state-of-the-art performance in a single run setting and converges faster than strong baselines. By mitigating the effects of spectral bias through frequency-guided alignment, SONAR provides a fully data-driven and architecture-agnostic approach to generalizable audio deepfake detection.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-hidekel26a, title = {{SONAR}: {S}pectral-{C}ontrastive Audio Residuals for Generalizable Deepfake Detection}, author = {Hidekel, Ido Nitzan and Lifshitz, Gal and Cohen, Khen and Raviv, Dan}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {42998--43012}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/hidekel26a/hidekel26a.pdf}, url = {https://proceedings.mlr.press/v306/hidekel26a.html}, abstract = {Deepfake audio detectors often fail to generalize to unseen attacks, in part due to spectral bias: neural networks prioritize low-frequency structure while under-exploiting subtle high-frequency (HF) artifacts left by generative models. We introduce SONAR (Spectral-cONtrastive Audio Residuals), a frequency-guided framework that explicitly enforces representation-level consistency between semantic content and HF residuals. Unlike prior frequency-aware or dual-stream detectors that treat HF cues as auxiliary features, SONAR encourages structured interaction between content and noise representations in latent space. The model employs a dual-path architecture in which an XLSR encoder captures low-frequency content, while a parallel branch with learnable, value-constrained 1D SRM (Spatial Rich Model) high-pass filters distills HF residuals. The two representations are fused via frequency cross-attention and trained with a Jensen–Shannon alignment loss that promotes LF–HF consistency for genuine audio and amplifies inconsistency for deepfakes. Evaluated on ASVspoof 2021 and in-the-wild benchmarks, SONAR achieves state-of-the-art performance in a single run setting and converges faster than strong baselines. By mitigating the effects of spectral bias through frequency-guided alignment, SONAR provides a fully data-driven and architecture-agnostic approach to generalizable audio deepfake detection.} }
Endnote
%0 Conference Paper %T SONAR: Spectral-Contrastive Audio Residuals for Generalizable Deepfake Detection %A Ido Nitzan Hidekel %A Gal Lifshitz %A Khen Cohen %A Dan Raviv %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-hidekel26a %I PMLR %P 42998--43012 %U https://proceedings.mlr.press/v306/hidekel26a.html %V 306 %X Deepfake audio detectors often fail to generalize to unseen attacks, in part due to spectral bias: neural networks prioritize low-frequency structure while under-exploiting subtle high-frequency (HF) artifacts left by generative models. We introduce SONAR (Spectral-cONtrastive Audio Residuals), a frequency-guided framework that explicitly enforces representation-level consistency between semantic content and HF residuals. Unlike prior frequency-aware or dual-stream detectors that treat HF cues as auxiliary features, SONAR encourages structured interaction between content and noise representations in latent space. The model employs a dual-path architecture in which an XLSR encoder captures low-frequency content, while a parallel branch with learnable, value-constrained 1D SRM (Spatial Rich Model) high-pass filters distills HF residuals. The two representations are fused via frequency cross-attention and trained with a Jensen–Shannon alignment loss that promotes LF–HF consistency for genuine audio and amplifies inconsistency for deepfakes. Evaluated on ASVspoof 2021 and in-the-wild benchmarks, SONAR achieves state-of-the-art performance in a single run setting and converges faster than strong baselines. By mitigating the effects of spectral bias through frequency-guided alignment, SONAR provides a fully data-driven and architecture-agnostic approach to generalizable audio deepfake detection.
APA
Hidekel, I.N., Lifshitz, G., Cohen, K. & Raviv, D.. (2026). SONAR: Spectral-Contrastive Audio Residuals for Generalizable Deepfake Detection. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:42998-43012 Available from https://proceedings.mlr.press/v306/hidekel26a.html.

Related Material