Reading Between the Waves: Robust Topic Segmentation Using Inter-Sentence Audio Features

TL;DR

This paper introduces a multi-modal approach combining text and acoustic features for robust topic segmentation in spoken content, significantly improving accuracy and noise resilience across multiple languages.

Contribution

It presents a novel multi-modal model that fine-tunes text and audio encoders to leverage acoustic cues, enhancing topic segmentation performance beyond existing text-only methods.

Findings

Significant improvements over text-only baselines on YouTube dataset

Enhanced robustness to ASR noise in multiple languages

Outperforms larger text-only models across diverse datasets

Abstract

Spoken content, such as online videos and podcasts, often spans multiple topics, which makes automatic topic segmentation essential for user navigation and downstream applications. However, current methods do not fully leverage acoustic features, leaving room for improvement. We propose a multi-modal approach that fine-tunes both a text encoder and a Siamese audio encoder, capturing acoustic cues around sentence boundaries. Experiments on a large-scale dataset of YouTube videos show substantial gains over text-only and multi-modal baselines. Our model also proves more resilient to ASR noise and outperforms a larger text-only baseline on three additional datasets in Portuguese, German, and English, underscoring the value of learned acoustic features for robust topic segmentation.