AHA: Aligning Large Audio-Language Models for Reasoning Hallucinations via Counterfactual Hard Negatives

Yanxi Chen, Wenhui Zhu, Xiwen Chen, Zhipeng Wang, Xin Li +7 more
12/30/2025
cs.SDcs.AIcs.CL

Official code: LLM-VLM-GSL/AHA (3 stars, last commit 2026-01-07).

Abstract

Although Large Audio-Language Models (LALMs) deliver state-of-the-art (SOTA) performance, they frequently suffer from hallucinations, e.g. generating text not grounded in the audio input. We analyze these grounding failures and identify a distinct taxonomy: Event Omission, False Event Identity, Temporal Relation Error, and Quantitative Temporal Error. To address this, we introduce the AHA (Audio Hallucination Alignment) framework. By leveraging counterfactual hard negative mining, our pipeline constructs a high-quality preference dataset that forces models to distinguish strict acoustic evidence from linguistically plausible fabrications. Additionally, we establish AHA-Eval, a diagnostic benchmark designed to rigorously test these fine-grained temporal reasoning capabilities. We apply this data to align Qwen2.5-Omni. The resulting model, Qwen-Audio-AHA, achieves a 13.7% improvement on AHA-Eval. Crucially, this benefit generalizes beyond our diagnostic set. Our model shows substantial gains on public benchmarks, including 1.3% on MMAU-Test and 1.6% on MMAR, outperforming latest SOTA methods. The model and dataset are open-sourced at https://github.com/LLM-VLM-GSL/AHA.

View on arXivView PDF

Code Implementations(1)

LLM-VLM-GSL/AHAOfficial100%

Code for Paper: AHA: Aligning Large Audio-Language Models for Reasoning Hallucinations via Counterfactual Hard Negatives

30PythonDec 30, 20259 months ago

Similar Papers

Reducing Object Hallucination in Large Audio-Language Models via Audio-Aware Decoding

86% match2025

Can Large Audio-Language Models Truly Hear? Tackling Hallucinations with Multi-Task Assessment and Stepwise Audio Reasoning

86% match1 repo342024

From Alignment to Advancement: Bootstrapping Audio-Language Alignment with Synthetic Data

85% match2025

ALARM: Audio-Language Alignment for Reasoning Models

85% match1 repo162026

Teaching Audio-Aware Large Language Models What Does Not Hear: Mitigating Hallucinations through Synthesized Negative Samples

85% match2025

MARCH: Multi-Agent Reinforced Self-Check for LLM Hallucination

84% match2026

Cite this paper

@article{chen2025aha,
  title  = {AHA: Aligning Large Audio-Language Models for Reasoning Hallucinations via Counterfactual Hard Negatives},
  author = {Yanxi Chen and Wenhui Zhu and Xiwen Chen and Zhipeng Wang and Xin Li and Peijie Qiu and Hao Wang and Xuanzhao Dong and Yujian Xiong and Anderson Schneider and Yuriy Nevmyvaka and Yalin Wang},
  year   = {2025},
  eprint = {2512.24052},
  archivePrefix = {arXiv},
  url    = {http://arxiv.org/abs/2512.24052v2}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/2512.24052.svg)](https://codeofpaper.com/paper/2512.24052)

Discussion