Explainable-by-Design Audio Deepfake Detection via Wiener-Hopf Linear Prediction

Authors: Mattia Tamiazzo, Simone Milani, Massimo Iuliani, Marco Fontani

Published: 2026-07-14 09:51:55+00:00

Comment: Accepted at ACM IH&MMSec 2026

AI Summary

This paper introduces an explainable-by-design audio deepfake detection framework that utilizes Wiener-Hopf linear prediction as a feature extraction method, followed by a lightweight 2D Convolutional Neural Network (CNN) for classification. This approach demonstrates competitive detection performance on benchmark datasets with significantly lower computational complexity than state-of-the-art solutions, while also providing interpretability through Grad-CAM analysis.

Abstract

The rapid advancement of synthetic speech generation methods has made audio deepfake detection a critical challenge in multimedia forensics. While recent approaches achieve high detection accuracy, they typically rely on black-box architectures that offer limited interpretability and high computational complexity. In this paper, we propose an explainable-by-design audio deepfake detection framework based on Wiener-Hopf linear prediction, processed by a lightweight 2D Convolutional Neural Network (CNN). This design enables a direct and transparent connection between classification outcomes and the acoustic properties of the signal. Experimental results on benchmark datasets demonstrate competitive detection performance while maintaining significantly lower computational complexity compared to state-of-the-art solutions. The interpretability analysis using Grad-CAM reveals that the classifier focuses on low-order predictor coefficients and on silence and transitional regions, suggesting that the Wiener-Hopf predictor captures reverberation characteristics and subtle statistical inconsistencies in synthetic speech. Finally, robustness experiments show that fine-tuning effectively recovers detection performance under common post-processing degradations, including additive noise, MP3 compression, and telephone filtering.

Key findings

The proposed framework achieves competitive detection performance (average EER of 3.16% and AUC of 0.987) with 20 times lower computational complexity compared to existing methods. Grad-CAM analysis revealed that the classifier focuses on low-order predictor coefficients and on silence and transitional regions, suggesting that the Wiener-Hopf predictor effectively captures reverberation characteristics and subtle statistical inconsistencies. Fine-tuning effectively recovers detection performance under various post-processing degradations like noise and compression.

Approach

The method involves extracting Wiener-Hopf linear prediction coefficients from audio segments, stacking them into a 2D matrix, and feeding this matrix into a lightweight 2D CNN for classification. The Wiener-Hopf coefficients are designed to capture acoustic properties related to reverberation and inconsistencies in synthetic speech, providing an explainable feature representation.

Datasets

ASVspoof 2019 (Logical Access scenario), FakeOrReal (FoR), Diffusion-Based Synthetic Speech Dataset (DiffSSD)

Model(s)

2D Convolutional Neural Network (CNN)

Author countries

Italy

← Previous