TAD: Token-Adaptive Contrastive Decoding with Confidence-Guided Gating for Hallucination Mitigation in Large Audio-Language Models

arXiv:2609.07286 · cs.SD, cs.CL, cs.CV · Submitted 2026-09-07 · Read on arXiv

cs.SD, cs.CL, cs.CV

Submitted: 2026-09-07

Updated: 2026-09-07

Comments: Accepted to Interspeech 2026

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Hallucination — 3× in this paper · explained in 3 episodes
• Hallucination Mitigation — 2× in this paper · explained in 2 episodes
• Adaptive Decoding — 1× in this paper · explained in 1 episode
• Audio-Aware Decoding — 1× in this paper
• Token-Adaptive Decoding — 1× in this paper

Related papers

• Few-Shot Open-Set Audio Classification via Transductive Prototype Refinement and Class Logit Enhancement
• Spectral Masking and Interpolation Attack (SMIA): A Black-box Adversarial Attack against Voice Authentication and Anti-Spoofing Systems
• AVMeme Exam: A Multimodal Multilingual Multicultural Benchmark for LLMs' Contextual and Cultural Knowledge and Thinking
• SoundWeaver: Compositional Warm-Starting for Text-to-Audio Diffusion Serving
• WASIL: In-the-Wild Arabic Spoken Interactions with LLMs
• Efficient Audiovisual Speech Processing via MUTUD: Multimodal Training and Unimodal Deployment