用思维链强化大音频模型,让自动分章更贴近编辑判断。
Listen, Reason, and Segment: Aligning LALMs with Editorial Judgment for Media Chapterization

- 通过思维链引导的组相对策略优化,对齐大音频模型与编辑决策。
- 相比最强基线,分章平均准确率提升49点,达新纪录。
- 适合媒体内容自动化处理、智能编目等实际场景使用。
大音频语言模型(LALMs)在标准评测上进展迅速,但在媒体工作流、内容整理和分发中的实际应用仍有限。我们识别出自动音频分章——将连续音频流划分为主题连贯章节——是一个高难度且具有商业价值的任务,暴露出当前模型的差距。该任务挑战在于章节边界依赖主观编辑判断,而非客观声学事件,要求模型在长音频上下文中进行序列推理,并逼近创作者设定的边界。我们提出 AudioChaps 框架,采用基于思维链(CoT)的组相对策略优化(GRPO)对端到端 LALM 进行后训练对齐。为支持训练与评估,我们构建了三个数据集:AudioChaps-Alignment(来自 YouTube 创作者标注的章节边界)、AudioChaps-CoT(提供结构化监督的高质量边界推理样本)、AudioChaps-Eval(保留测试基准)。无需监督微调冷启动,AudioChaps-R1-Zero 已使平均 F1 提升 33 点,超越当前最佳模型 Audio-Flamingo-3-Think。最终模型 AudioChaps-R1 再提升 49 点,证明 GRPO 训练的 LALM 可可靠地将非结构化音频转化为可导航的结构化媒体。代码、模型与数据集将在接受后公开于 https://github.com/ta012/AudioChaps。
原文摘要 · Abstract (English)
Large Audio Language Models (LALMs) have made rapid progress on standardized benchmarks, yet their deployment in practical media workflows, curation, archival indexing, and content distribution remains largely unrealized. We identify automated audio chapterization, the task of segmenting continuous audio streams into thematically coherent chapters, as a demanding and commercially consequential setting that exposes this gap. Chapterization is challenging because boundaries are defined less by objective acoustic events than by subjective editorial judgment, requiring models to reason sequentially over long acoustic contexts and approximate creator-authored boundary decisions. We present AudioChaps, a post-training framework for aligning end-to-end LALMs for this task via Group Relative Policy Optimization (GRPO) guided by Chain-of-Thought (CoT) reasoning. To support training and evaluation, we curate three datasets: AudioChaps-Alignment, derived from creator-annotated chapter boundaries on YouTube; AudioChaps-CoT, which provides structured supervision for well-formatted, high-quality, and evidence-grounded boundary reasoning; and AudioChaps-Eval, a held-out benchmark for audio chapterization. Applying GRPO directly without a Supervised Fine-Tuning (SFT) cold start, AudioChaps-R1-Zero already improves average F1 by 33 points over the state-of-the-art LALM Audio-Flamingo-3-Think. The AudioChaps framework produces our final aligned LALM, AudioChaps-R1, which improves average F1 by 49 points. These results demonstrate that GRPO-trained LALMs can reliably transform unstructured auditory streams into navigable, structured media. Our code, models, and dataset resources will be released upon acceptance at https://github.com/ta012/AudioChaps.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品;中文卡片由大模型生成,请以原文为准。