Listen, Reason, and Segment: Aligning LALMs with Editorial Judgment for Media Chapterization

Excerpt

arXiv:2608.16539v1 Announce Type: cross Abstract: 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 chal