cs.SDNov 11, 2026

DeALR: Decoupled Audio-Latent Optimization with Acoustic Backtracking for Test-Time Reasoning

Authors: Haoyu He, Yan Rong, Li Liu

Organizations: HKUST(GZ)

Abstract

Large audio-language models (LALMs) typically encode audio once and conduct subsequent multi-step reasoning in a language-dominated space, weakening access to fine-grained acoustic evidence. Although recent multimodal approaches have explored re-engaging perceptual inputs during reasoning, whether a frozen LALM can actively revisit crucial acoustic evidence in the latent space at test time remains underexplored. To bridge this gap, we introduce Decoupled Audio Latent Reasoning (DeALR), a novel test-time latent reasoning framework that shifts the paradigm from text-dominated deliberation to active audio-latent interaction. Inspired by the human cognitive process of "backtracking" to specific acoustic cues during complex listening tasks, DeALR establishes a dynamic interaction between learnable thought tokens and acoustic segments. Specifically, we introduce Span-Level Acoustic Backtracking, which traces initial attention distributions to dynamically isolate contiguous, key acoustic spans. Furthermore, we propose Decoupled Audio-Latent Optimization (DALO), a test-time reinforcement learning algorithm that jointly refines thought vectors and acoustic patches via confidence-guided policy gradients. Crucially, DALO explicitly decouples trajectory search from response generation by reinstating uncorrupted, high-fidelity audio embeddings prior to text synthesis, entirely eliminating optimization-induced distribution drift while retaining the navigation of the refined thoughts. Comprehensive experiments on benchmarks demonstrate that DeALR consistently outperforms prior test-time optimization approaches, delivering substantial improvements in both confidence calibration and reasoning accuracy.

Figures & tables

Explore similar work

Sep 20, 2026

Listen Then Reason: Perception-Grounded Test-Time Reinforcement Learning for Large Audio-Language Models

Large audio-language models (LALMs) are increasingly used for a broader range of audio reasoning tasks. These models typically incorporate audio representations into a large language model (LLM) backbone to enable multimodal reasoning. Recent test-time reinforcement learning (TTRL) methods further improve LLM reasoning capability by leveraging unlabelled test data after pre-training. However, the importance of the perceptual capability of LALMs remains underexplored, particularly how much acoustic evidence is integrated and relied upon during reasoning, and how this contributes to final task performance. This gap limits the development of effective post-training methods like TTRL for audio reasoning. In this work, we first analyse how audio information is integrated and utilised during reasoning process. We quantify layer-wise perceptual reliance and show that stronger acoustic reliance is associated with higher accuracy and a larger performance gain attributable to the audio input. Building on this, we propose Perception-Grounded TTRL (PG-TTRL), which aligns label-free test-time optimisation with perceptually grounded reasoning, encouraging the model to structure its reasoning more strongly on the audio input. Experiments across LALMs and benchmarks show that PG-TTRL consistently improves reasoning performance over both the base models and standard TTRL, showing the value of perceptual-grounding optimisation for test-time audio reasoning.
Apr 20, 2026

Audio-DeepThinker: Progressive Reasoning-Aware Reinforcement Learning for High-Quality Chain-of-Thought Emergence in Audio Language Models

Large Audio-Language Models (LALMs) have made significant progress in audio understanding, yet they primarily operate as perception-and-answer systems without explicit reasoning processes. Existing methods for enhancing audio reasoning rely either on supervised chain-of-thought (CoT) fine-tuning, which is limited by training data quality, or on reinforcement learning (RL) with coarse rewards that do not directly evaluate reasoning quality. As a result, the generated reasoning chains often appear well-structured yet lack specific acoustic grounding. We propose Audio-DeepThinker, a framework built on two core ideas. First, we introduce a hybrid reasoning similarity reward that directly supervises the quality of generated reasoning chains by combining an LLM evaluator assessing logical path alignment, key step coverage, and analytical depth with an embedding similarity component enforcing semantic alignment with reference reasoning chains. Second, we propose a progressive two-stage curriculum that enables high-quality CoT reasoning to emerge through pure RL exploration, without any supervised reasoning fine-tuning, from an instruction-tuned model that possesses no prior chain-of-thought capability. Stage 1 trains on foundational audio QA with the hybrid reward to foster basic reasoning patterns, while Stage 2 shifts to acoustically challenging boundary cases with an LLM-only reward for greater reasoning diversity. Audio-DeepThinker achieves state-of-the-art results on MMAR (74.0%), MMAU-test-mini (78.5%), and MMSU (77.26%), winning 1st Place in the Interspeech 2026 Audio Reasoning Challenge (Single Model Track). Interpretability analyses further reveal that RL training primarily reshapes upper-layer MoE gating mechanisms and that reasoning tokens crystallize progressively in the upper transformer layers, offering mechanistic insights into how audio reasoning emerges through exploration.
Oct 17, 2026

Long Listening Thoughts: Eliciting Open Auditory Reasoning with Deliberative Perception and Cognitive Refinement

We introduce Long Listening Thoughts (LLT), a framework for constructing post-training datasets that improves audio reasoning by pairing deliberate perceptual grounding with explicit cognitive behaviors. On audio understanding and reasoning benchmarks, large audio-language models (LALMs) are limited primarily by perceptual errors, yet existing post-training datasets rely on audio descriptions of limited diversity. Through a four-stage pipeline, LLT (1) builds multi-aspect dense audio descriptions spanning semantic, acoustic, and temporal information, (2) generates more challenging questions, (3) synthesizes incorrect reasoning traces, and (4) extends short reasoning traces into long ones via a thought-continuation strategy that elicits cognitive behaviors such as verification and backtracking. Using LLT, we synthesize 1.6M reasoning traces and train Qwen2.5-Omni-7B with a two-stage pipeline of supervised fine-tuning (SFT) followed by direct preference optimization (DPO). This improves accuracy on MMAU Test-mini (71.5 to 78.2), MMAR (56.7 to 63.7), and MMSU (60.6 to 66.0).