Toward Interpretable Speech Deepfake Detection using Artifact-Specific Experts and Calibrated Detection Scores

2026-07-23

AI Summary

This work addresses the limited interpretability of existing deepfake audio detection methods, which predominantly rely on black-box models and thus fail to meet trustworthiness requirements in high-stakes scenarios. To overcome this limitation, the authors propose an interpretable detection framework based on artifact-specific experts: each expert specializes in identifying a distinct type of synthetic artifact and outputs a probability-calibrated log-likelihood ratio as human-understandable evidence. These expert predictions are then aggregated through ensemble learning to produce the final decision. Experimental results demonstrate that the proposed approach not only effectively captures interpretable artifact signals introduced by diverse generative models but also achieves competitive detection performance while significantly enhancing decision transparency and system trustworthiness.

Abstract

In this work, we propose an interpretable framework for speech deepfake detection based on artifact-specific expert models. Rather than relying on black-box decisions, the framework provides human-understandable evidence, which is critical in high-stakes settings. Each expert is trained to detect a specific speech synthesis artifact, and its output is calibrated into a log-likelihood ratio that serves as an interpretable evidence score. We evaluate five artifact-specific experts and show that, with proper calibration, they can capture their target artifacts and produce meaningful evidence. Importantly, each expert estimates only the presence of its assigned artifact rather than directly performing the final decision. Their outputs are aggregated into an ensemble to produce the actual real-versus-fake classification, while maintaining interpretability by indicating how strongly each expert supports or contradicts a fake classification. Results show that artifact-specific experts capture interpretable signals of synthetic speech across multiple generation pipelines.

Problem

• speech deepfake detection
• interpretability
• artifact-specific
• calibrated scores
• synthetic speech

Innovation

• interpretable detection
• artifact-specific experts
• speech deepfake
• calibrated scores
• log-likelihood ratio