When Scaling Fails: Mitigating Audio Perception Decay of LALMs via Multi-Step Perception-Aware Reasoning
cs.SD, cs.AI, eess.AS
Submitted: 2026-02-28
Updated: 2026-09-21
Comments: Accepted by EMNLP 2026 Main Conference
Code: https://github.com/Moriiikdt/MPAR2
License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
Terminology
Sources
- Audio Flamingo 3: Advancing Audio Intelligence with Fully Open Large Audio Language Models
- DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning
- GPT-4o System Card
- OpenAI o1 System Card
- Evaluating Step-by-step Reasoning Traces: A Survey
- Phi-4-Mini Technical Report: Compact yet Powerful Multimodal Language Models via Mixture-of-LoRAs
- Probing the Difficulty Perception Mechanism of Large Language Models
- Reinforcement Learning Outperforms Supervised Fine-Tuning: A Case Study on Audio Question Answering
- Large Language Monkeys: Scaling Inference Compute with Repeated Sampling
- DeSTA2.5-Audio: Toward General-Purpose Large Audio Language Model with Self-Generated Cross-Modal Alignment
- Efficient Prompting Methods for Large Language Models: A Survey
- Audio-CoT: Exploring Chain-of-Thought Reasoning in Large Audio Language Model
- Qwen2-Audio Technical Report
- MMAR: A Challenging Benchmark for Deep Reasoning in Speech, Audio, Music, and Their Mix
- Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilities
- DeepSeek-R1 Thoughtology: Let's think about LLM Reasoning
- S-GRPO: Early Exit via Reinforcement Learning in Reasoning Models
- Qwen2.5-Omni Technical Report
- Omni-R1: Do You Really Need Audio to Fine-Tune Your Audio LLM?
- Qwen3 Technical Report
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