SVHalluc: Benchmarking Speech-Vision Hallucination in Audio-Visual Large Language Models

• URL: http://arxiv.org/abs/2606.02642v1
• Date: Sun, 31 May 2026 14:00:37 GMT
• Title: SVHalluc: Benchmarking Speech-Vision Hallucination in Audio-Visual Large Language Models
• Abstract summary: We show that speech content can induce hallucinations in audio-visual large-language models (LLMs)<n>Our benchmark diagnoses speech-vision hallucinations from two critical and complementary aspects: semantic and temporal.<n>Our work uncovers a new and fundamental limitation of current audio-visual LLMs and highlights the need for speech-grounded video comprehension.
• Score: 32.25458333197992
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Despite the success of audio-visual large-language models (LLMs), they can produce plausible but ungrounded outputs, termed hallucination. Existing benchmarks focus on environmental sounds (e.g., dog barking) to indicate event occurrence. In contrast, human speech carries fundamentally different, rich semantics and temporal structures, yet it remains unexplored whether current models can accurately align speech content with corresponding visual signals. In this work, we show that speech content can induce hallucinations in audio-visual LLMs. To systematically study this, we introduce SVHalluc, the first comprehensive benchmark for evaluating speech-vision hallucination in audio-visual LLMs. Our benchmark diagnoses speech-vision hallucinations from two critical and complementary aspects: semantic and temporal. Experimental results demonstrate that state-of-the-art open-source audio-visual LLMs struggle with aligning speech content with corresponding visual signals, with a near-random accuracy on multiple tasks. In contrast, Gemini 2.5 Pro significantly outperforms the open-source models. Our analysis suggests that their failures stem from limited ability in cross-modality understanding, despite strong performance in single-modality perception. Our work uncovers a new and fundamental limitation of current audio-visual LLMs and highlights the need for speech-grounded video comprehension. Project page: https://chenshuang-zhang.github.io/projects/svhalluc/.

Related papers

• Do Audio-Visual Large Language Models Really See and Hear? [66.8531220331223]
  We analyze how audio and visual features evolve and fuse through different layers of an AVLLM to produce the final text outputs.<n>Our findings reveal a fundamental modality bias in AVLLMs and provide new mechanistic insights into how multimodal LLMs integrate audio and vision.
  arXiv  Detail & Related papers  (2026-04-03T00:48:49Z)
• See, Hear, and Understand: Benchmarking Audiovisual Human Speech Understanding in Multimodal Large Language Models [24.851643680674474]
  AV-SpeakerBench is a curated benchmark of 3,212 multiple-choice questions focused on speaker-centric audiovisual reasoning in real-world videos.<n>It features: (1) a speaker-centered formulation that treats speakers-not scenes-as the core reasoning unit; (2) fusion-grounded question design embedding audiovisual dependencies into question semantics; and (3) expert-curated annotations ensuring temporal precision and cross-modal validity.
  arXiv  Detail & Related papers  (2025-12-01T21:57:26Z)
• When Eyes and Ears Disagree: Can MLLMs Discern Audio-Visual Confusion? [41.579901082251254]
  Experimental results reveal that Multimodal Large Language Models (MLLMs) struggle to discriminate non-existent audio due to visually dominated reasoning.<n>Motivated by this observation, we introduce RL-CoMM, a Reinforcement Learning-based Collaborative Multi-MLLM.<n> RL-CoMM includes two stages: 1) To alleviate visually dominated ambiguities, we introduce an external model, a Large Audio Language Model (LALM), as the reference model to generate audio-only reasoning.
  arXiv  Detail & Related papers  (2025-11-13T07:59:41Z)
• Bridging Ears and Eyes: Analyzing Audio and Visual Large Language Models to Humans in Visible Sound Recognition and Reducing Their Sensory Gap via Cross-Modal Distillation [13.137446396934102]
  We evaluate audio, visual, and audio-visual large language models (LLMs) against humans in recognizing sound objects.<n>We uncover a performance gap between Qwen2-Audio and Qwen2-VL that parallels the sensory discrepancy between human ears and eyes.
  arXiv  Detail & Related papers  (2025-05-11T01:01:44Z)
• AVHBench: A Cross-Modal Hallucination Benchmark for Audio-Visual Large Language Models [27.430040932849018]
  We introduce AVHBench, the first comprehensive benchmark specifically designed to evaluate the perception and comprehension capabilities of audio-visual models.<n>Our results reveal that most existing audio-visual LLMs struggle with hallucinations caused by cross-interactions between modalities.<n>We demonstrate that simple training with our AVHBench improves robustness of audio-visual LLMs against hallucinations.
  arXiv  Detail & Related papers  (2024-10-23T23:36:06Z)
• Speech2rtMRI: Speech-Guided Diffusion Model for Real-time MRI Video of the Vocal Tract during Speech [29.510756530126837]
  We introduce a data-driven method to visually represent articulator motion in MRI videos of the human vocal tract during speech. We leverage large pre-trained speech models, which are embedded with prior knowledge, to generalize the visual domain to unseen data.
  arXiv  Detail & Related papers  (2024-09-23T20:19:24Z)
• Hallucination Augmented Contrastive Learning for Multimodal Large Language Model [53.65682783591723]
  Multi-modal large language models (MLLMs) have been shown to efficiently integrate natural language with visual information to handle multi-modal tasks. However, MLLMs still face a fundamental limitation of hallucinations, where they tend to generate erroneous or fabricated information. In this paper, we address hallucinations in MLLMs from a novel perspective of representation learning.
  arXiv  Detail & Related papers  (2023-12-12T04:05:15Z)
• LA-VocE: Low-SNR Audio-visual Speech Enhancement using Neural Vocoders [53.30016986953206]
  We propose LA-VocE, a new two-stage approach that predicts mel-spectrograms from noisy audio-visual speech via a transformer-based architecture. We train and evaluate our framework on thousands of speakers and 11+ different languages, and study our model's ability to adapt to different levels of background noise and speech interference.
  arXiv  Detail & Related papers  (2022-11-20T15:27:55Z)
• Audio-Visual Speech Codecs: Rethinking Audio-Visual Speech Enhancement by Re-Synthesis [67.73554826428762]
  We propose a novel audio-visual speech enhancement framework for high-fidelity telecommunications in AR/VR. Our approach leverages audio-visual speech cues to generate the codes of a neural speech, enabling efficient synthesis of clean, realistic speech from noisy signals.
  arXiv  Detail & Related papers  (2022-03-31T17:57:10Z)
• LiRA: Learning Visual Speech Representations from Audio through Self-supervision [53.18768477520411]
  We propose Learning visual speech Representations from Audio via self-supervision (LiRA) Specifically, we train a ResNet+Conformer model to predict acoustic features from unlabelled visual speech. We show that our approach significantly outperforms other self-supervised methods on the Lip Reading in the Wild dataset.
  arXiv  Detail & Related papers  (2021-06-16T23:20:06Z)
• "Notic My Speech" -- Blending Speech Patterns With Multimedia [65.91370924641862]
  We propose a view-temporal attention mechanism to model both the view dependence and the visemic importance in speech recognition and understanding. Our proposed method outperformed the existing work by 4.99% in terms of the viseme error rate. We show that there is a strong correlation between our model's understanding of multi-view speech and the human perception.
  arXiv  Detail & Related papers  (2020-06-12T06:51:55Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.