PaperScope
LIVE · 2026-09-21 05:40 UTC

Reusing Latent Speech Representations for Query-Conditioned Topic Localization in Transcripts

Steffen Freisinger, Philipp Seeberger, Thomas Ranzenberger, Tobias Bocklet, Korbinian Riedhammer

Latestcs.CLcs.LGcs.AIcs.CV
arXiv ID
2609.21844 v1
Category
Submitted
2026-09-18

Abstract

Long transcripts are costly inputs for downstream NLP systems and often contain irrelevant context. We study query-conditioned topic localization: predicting the sentence span in a transcript that best addresses a topic-title query. To improve span localization, we reuse ASR encoder states as sentence-level representations and fuse them with textual embeddings. This lets lightweight span locators exploit speech information without running a separate audio encoder. Experiments on two public datasets show consistent gains over text-only baselines, especially under strict boundary-matching criteria. Cross-dataset experiments further indicate that the benefits are strongest for structured or semi-structured speech, while gains on spontaneous speech are limited and mixed.

Comment: Accepted at EMNLP 2026 Main Conference

arXiv abs page · PDF · same-day batch