SE-ADD: Self-Evolving Audio Deepfake Detection with Mistake-Driven Supervision

2026-09-30

AI Summary

This study addresses the limited post-deployment generalization of audio deepfake detection systems against unknown attacks and their neglect of model misjudgments. To overcome these limitations, this work proposes a self-evolving framework built upon audio language models. It introduces a novel error-driven supervision mechanism that leverages model misclassifications to automatically generate forensic cues, thereby constructing an evolving spoofing environment. Through iterative Low-Rank Adaptation (LoRA), the framework achieves adaptive continual learning without requiring manual annotations. Experimental results demonstrate that the proposed method significantly enhances defensive capabilities against unseen attacks, reducing the equal error rates of Qwen2-Audio and MOSS-Audio to 7.52% and 3.97%, respectively.

Abstract

Audio deepfake detection (ADD) must remain effective when new spoofing attacks emerge after deployment. Emerging audio language model (ALM)-based ADD methods are built on predefined supervision from ground-truth labels or verified forensic rationales. However, this paradigm overlooks an ALM's own mistakes, which indicate where targeted supervision is most needed. To this end, we first introduce evolving spoofing environments for ALM-based ADD, where a new attack becomes dominant while previously observed attacks persist. Motivated by the above learning-from-mistakes perspective, we further propose SE-ADD, a self-evolving framework that iteratively adapts an ALM via low-rank adaptation (LoRA) using mistake-driven supervision built from its verdicts and self-generated forensic cues. All training samples receive direct authenticity supervision, while misclassified ones receive additional cue-augmented supervision. As verdicts and cues are regenerated by the updated ALM, the resulting supervision evolves accordingly. Experiments on two ALMs demonstrate the effectiveness of SE-ADD in generalizing to unseen attacks, reducing the equal error rate (EER) from $36.72\%$ to $7.52\%$ for Qwen2-Audio and from $19.93\%$ to $3.97\%$ for MOSS-Audio.

Problem

• Audio Deepfake Detection
• Self-Evolving
• Audio Language Model
• Evolving Spoofing Environments
• Unseen Attacks

Innovation

• Self-Evolving Framework
• Mistake-Driven Supervision
• Audio Deepfake Detection
• Low-Rank Adaptation
• Audio Language Model