SWEAR: Section-Wise Efficient Audio Reasoning via Self-Rated Difficulty

arXiv:2610.13081v1 Announce Type: new Abstract: Recent work on large audio-language models (LALMs) explores adaptive reasoning to balance accuracy and efficiency. However, existing approaches often rely on coarse-grained difficulty estimation, overlooking the diverse factors that determine reasoning effort. We address this limitation by deriving five-dimensional difficulty scores and compressed reasoning traces from complementary offline assessments. Based on this characterization, we propose SWEAR, a difficulty-aware reasoning framework that learns difficulty signals during supervised fine-tuning and uses reinforcement learning to adapt reasoning allocation across reasoning stages. Experiments on MMAU-test-mini and MMAR show that SWEAR improves accuracy while substantially reducing generated tokens, shortening reasoning for easy instances without sacrificing performance on difficult ones.

Related

• Audio-Zero: Label-Free Self-Evolution for Fine-Grained Audio Reasoning
• SLAT: Segment-Level Adaptive Trimming for Efficient CoT Reasoning
• Dual-Dimensional Consistency: Balancing Budget and Quality in Adaptive Inference-Time Scaling
• Adaptive Perturbation Selection for Contrastive Audio Decoding

Source: arXiv cs.AI | 2026-10-12