# Large Audio Language Models

Audio language model benchmark scores can mistake language ability for acoustic understanding. Using synthetic readings and transcripts as references, the study measures the recording gain of the original audio over words across 92 tasks from 6 benchmarks, finding that 19 open models trail Gemini 3.1 Pro by 28.6 and 41.9 percentage points on global-attribute and within-recording structure tasks, with about half or more of each gap lying in the recording gain. Training LoRA adapters with a content anchor in the readout-decline layers of Gemma-4 12B and Qwen2.5-Omni-7B raises attribute and structure accuracy by 12.8 to 44.4 points while mean lexical accuracy stays within 1 point of the base models.

Audio language model benchmark scores can mistake language ability for acoustic understanding. Using synthetic readings and transcripts as references, the study measures the recording gain of the original audio over words across 92 tasks from 6 benchmarks, finding that 19 open models trail Gemini 3.1 Pro by 28.6 and 41.9 percentage points on global-attribute and within-recording structure tasks, with about half or more of each gap lying in the recording gain. Training LoRA adapters with a content anchor in the readout-decline layers of Gemma-4 12B and Qwen2.5-Omni-7B raises attribute and structure accuracy by 12.8 to 44.4 points while mean lexical accuracy stays within 1 point of the base models.

Large audio-language models that reason explicitly have mostly used one overall difficulty signal to decide whether to think longer, which gives little guidance on how much effort each part of the reasoning trace deserves. SWEAR scores difficulty along five dimensions covering acoustic quality, acoustic complexity and reasoning demand, uses those offline scores to compress reference traces, teaches the model during supervised fine-tuning to predict each section's difficulty before writing it, and then lets reinforcement learning set section length from those self-ratings. On MMAU-test-mini and MMAR, accuracy rises by 2.6 and 6.3 percentage points over CoTA-SFT without difficulty supervision, while generated tokens fall by 60.1% and 62.3%.

Large audio-language models that reason explicitly have mostly used one overall difficulty signal to decide whether to think longer, which gives little guidance on how much effort each part of the reasoning trace deserves. SWEAR scores difficulty along five dimensions covering acoustic quality, acoustic complexity and reasoning demand, uses those offline scores to compress reference traces, teaches the model during supervised fine-tuning to predict each section's difficulty before writing it, and then lets reinforcement learning set section length from those self-ratings. On MMAU-test-mini and MMAR, accuracy rises by 2.6 and 6.3 percentage points over CoTA-SFT without difficulty supervision, while generated tokens fall by 60.1% and 62.3%.

Deleting the waveform region an audio-language model's claim points to overturns its "sound present" answer more often than deleting an equally sized random window, indicating that these hallucinations are at least partly grounded in locatable audio evidence, and the same contrast lifts Qwen2-Audio accuracy from 57.1% to 71.9%.

Deleting the waveform region an audio-language model's claim points to overturns its "sound present" answer more often than deleting an equally sized random window, indicating that these hallucinations are at least partly grounded in locatable audio evidence, and the same contrast lifts Qwen2-Audio accuracy from 57.1% to 71.9%.

AdaLoop replaces the linear projection between an audio encoder and a language model with a lightweight recurrent module in which a shared, question-guided transformer block iteratively refines the audio representation and a learned halting mechanism decides how many steps each audio–question pair needs; across Qwen2.5-Omni-7B, Kimi-Audio-7B-Instruct, and MiMo-Audio-7B-Instruct on MMSU, MMAU-Pro, and MMAR it raises average accuracy by 2.9 to 3.8 points, with gains concentrated on perception-heavy subtasks, and the learned depth distribution shows signal-level questions averaging 5.7 iterations versus 2.6 for semantic ones.

AdaLoop replaces the linear projection between an audio encoder and a language model with a lightweight recurrent module in which a shared, question-guided transformer block iteratively refines the audio representation and a learned halting mechanism decides how many steps each audio–question pair needs; across Qwen2.5-Omni-7B, Kimi-Audio-7B-Instruct, and MiMo-Audio-7B-Instruct on MMSU, MMAU-Pro, and MMAR it raises average accuracy by 2.9 to 3.8 points, with gains concentrated on perception-heavy subtasks, and the learned depth distribution shows signal-level questions averaging 5.7 iterations versus 2.6 for semantic ones.

The work proposes a framework that fine-tunes Large Audio Language Models (LALMs) for dysarthric speech detection on speech recordings combined with textual information comprising low-level acoustic features and speaker demographics; across two LALMs the framework outperforms deep-learning-based baselines, with Qwen2-Audio-Instruct achieving state-of-the-art performance, and an ablation study shows that incorporating acoustic features and speaker demographics during fine-tuning improves LALM performance, while LALMs alone exhibit only chance-level zero-shot performance.

The work proposes a framework that fine-tunes Large Audio Language Models (LALMs) for dysarthric speech detection on speech recordings combined with textual information comprising low-level acoustic features and speaker demographics; across two LALMs the framework outperforms deep-learning-based baselines, with Qwen2-Audio-Instruct achieving state-of-the-art performance, and an ablation study shows that incorporating acoustic features and speaker demographics during fine-tuning improves LALM performance, while LALMs alone exhibit only chance-level zero-shot performance.

This work ports multimodal Direct Preference Optimization (mDPO) from vision-language models to the audio domain, building preference pairs that contrast intact audio with acoustically perturbed audio (reversal, random noise, time and frequency masking) and training Qwen2-Audio-7B-Instruct, raising DCASE 2025 audio QA accuracy from 56.0% to 70.0% and AH Existence accuracy from 55.6% to 83.0%.

This work ports multimodal Direct Preference Optimization (mDPO) from vision-language models to the audio domain, building preference pairs that contrast intact audio with acoustically perturbed audio (reversal, random noise, time and frequency masking) and training Qwen2-Audio-7B-Instruct, raising DCASE 2025 audio QA accuracy from 56.0% to 70.0% and AH Existence accuracy from 55.6% to 83.0%.