SEAR: Spoofing Evidence-Grounded Audio Reasoning Benchmark for Audio Language Models

2026-09-30

AI Summary

This study addresses the limitation of existing audio deepfake detection benchmarks, which evaluate only the plausibility of conclusions or rationales without verifying whether models genuinely reason from acoustic evidence. To bridge this gap, this work introduces the paradigm of evidence-grounded audio reasoning, proposing the SEAR benchmark and the BAEA agent. Through a four-task Audio Question Answering (AQA) design, a frozen-model architecture augmented with external acoustic tools, and fixed or adaptive evidence acquisition strategies, the proposed framework compels models to base their detection on verifiable acoustic evidence. Experimental results demonstrate that BAEA-Fixed substantially improves both verdict accuracy and forensic quality. Furthermore, it reveals a critical discrepancy between plausible rationales and authentic evidence, confirming that misleading evidence can severely degrade detection performance.

Abstract

Audio language models (ALMs) are increasingly used for audio deepfake detection (ADD), yet existing benchmarks assess their verdicts or rationale plausibility without verifying the underlying acoustic evidence. To address this issue, we first introduce spoofing evidence-grounded audio reasoning (SEAR), a four-task AQA benchmark to evaluate ALM-based ADD through acoustic evidence identification and quantification, deepfake detection, and forensic rationale generation. We further propose a bona-fide-based acoustic evidence agent (BAEA), which equips a frozen ALM with controlled acoustic tools under \textsc{fixed} or \textsc{adaptive} evidence-acquisition policies. Experiments with six ALMs reveal a clear gap between plausible rationales and verifiable acoustic evidence reasoning, while BAEA-\textsc{Fixed} improves final verdicts and forensic rationales on both evaluation partitions. Controlled interventions further show that misleading evidence degrades both detection and grounding performance.

Problem

• Audio Deepfake Detection
• Audio Language Models
• Acoustic Evidence Grounding
• Benchmark Evaluation

Innovation

• Audio Language Models
• Audio Deepfake Detection
• Acoustic Evidence Reasoning
• Evidence Agent
• Benchmark