# Does Location Matter? A Black-Box Causal Audit Reveals Model-Dependent Audio Grounding in LALM Hallucination

arXiv: https://arxiv.org/abs/2610.13112 · HTML: https://mlsanity.com/paper/2610.13112/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 20 of its 24 references that are in the map.
- First posted on arXiv 9 Oct 2026.

## Builds on (most cited first, 20 of 20)

- [RISE: Randomized Input Sampling for Explanation of Black-box Models](https://mlsanity.com/paper/1806.07421/) (2018)
- [Large-scale Contrastive Language-Audio Pretraining with Feature Fusion and Keyword-to-Caption Augmentation](https://mlsanity.com/paper/2211.06687/) (2022)
- [Mitigating Object Hallucinations in Large Vision-Language Models through Visual Contrastive Decoding](https://mlsanity.com/paper/2311.16922/) (2023)
- [Contrastive Decoding: Open-ended Text Generation as Optimization](https://mlsanity.com/paper/2210.15097/) (2022)
- [Qwen2-Audio Technical Report](https://mlsanity.com/paper/2407.10759/) (2024)
- [Understanding Deep Networks via Extremal Perturbations and Smooth Masks](https://mlsanity.com/paper/1910.08485/) (2019)
- [Audio Flamingo 3: Advancing Audio Intelligence with Fully Open Large Audio Language Models](https://mlsanity.com/paper/2507.08128/) (2025)
- [Multi-Modal Hallucination Control by Visual Information Grounding](https://mlsanity.com/paper/2403.14003/) (2024)
- [Mitigating Hallucinations in Large Vision-Language Models with Instruction Contrastive Decoding](https://mlsanity.com/paper/2403.18715/) (2024)
- [Probabilities of Causation: Bounds and Identification](https://mlsanity.com/paper/1301.3898/) (2013)
- [Can Large Audio-Language Models Truly Hear? Tackling Hallucinations with Multi-Task Assessment and Stepwise Audio Reasoning](https://mlsanity.com/paper/2410.16130/) (2024)
- [Reducing Object Hallucination in Large Audio-Language Models via Audio-Aware Decoding](https://mlsanity.com/paper/2506.07233/) (2025)
- [FLAM: Frame-Wise Language-Audio Modeling](https://mlsanity.com/paper/2505.05335/) (2025)
- [HalluAudio: A Comprehensive Benchmark for Hallucination Detection in Large Audio-Language Models](https://mlsanity.com/paper/2604.19300/) (2026)
- [Causal Tracing of Audio-Text Fusion in Large Audio Language Models](https://mlsanity.com/paper/2603.13768/) (2026)
- [Temporal Contrastive Decoding: A Training-Free Method for Large Audio-Language Models](https://mlsanity.com/paper/2604.15383/) (2026)
- [COPO: Causal-Oriented Policy Optimization for Hallucinations of MLLMs](https://mlsanity.com/paper/2508.04182/) (2025)
- [Encode Once, Decode Never: Reusing Audio LM Internals for Efficient Temporal Localization](https://mlsanity.com/paper/2602.10230/) (2026)
- [Tracing Audio Grounding and Answer Selection in Audio LLMs](https://mlsanity.com/paper/2609.04637/) (2026)
- [TAD: Token-Adaptive Contrastive Decoding with Confidence-Guided Gating for Hallucination Mitigation in Large Audio-Language Models](https://mlsanity.com/paper/2609.07286/) (2026)

## Abstract (arXiv)

Large audio-language models (LALMs) frequently exhibit object hallucination, affirming absent sounds. The dominant mitigation, Audio-Aware Decoding (AAD), contrasts logits under real audio against an all-zero "blank" waveform, flagged by prior review as out-of-distribution. We introduce a black-box causal test that isolates the claimed evidence itself: localize a claim's evidence region via CLAP similarity, remove it by spectral gating, and compare against a random-window control across a 2x2x2 grid on three LALMs. On Qwen2-Audio-7B, localized ablation's raw AUROC trails blank's (0.613 vs. 0.721); this is a duration confound, not a localization failure, since blank removes the entire clip versus the grid's 2-4 s. Holding duration fixed, localization beats random removal at every cell on both models. Raw detection value remains model-dependent, especially on Audio Flamingo 3. The same evidence corrects decoding: contrastive decoding combining blank and language-prior contrasts, gated and ensembled, raises Qwen2-Audio accuracy from 57.1% to 71.9%, beating AAD alone. A unified causal-grounding profile does not transfer between models, indicating different grounding regimes. The source code and a detailed report of our findings, including identified limitations, are available at https://github.com/khalooei/causa.