Topic Model Robustness to Automatic Speech Recognition Errors in Podcast Transcripts

TL;DR

This paper investigates how robust Latent Dirichlet Allocation topic models are when applied to automatic speech recognition transcripts of Danish podcasts, showing that high-quality topics can still be extracted despite transcription errors.

Contribution

It demonstrates the resilience of LDA topic modeling to ASR errors in low-resource language transcripts, providing insights for improved content relevance in multilingual podcast services.

Findings

Topic embeddings remain high quality despite increasing transcription noise.

Cosine similarity scores decrease with more ASR errors but stay informative.

Robustness of LDA supports use of automatic transcripts for content analysis.

Abstract

For a multilingual podcast streaming service, it is critical to be able to deliver relevant content to all users independent of language. Podcast content relevance is conventionally determined using various metadata sources. However, with the increasing quality of speech recognition in many languages, utilizing automatic transcriptions to provide better content recommendations becomes possible. In this work, we explore the robustness of a Latent Dirichlet Allocation topic model when applied to transcripts created by an automatic speech recognition engine. Specifically, we explore how increasing transcription noise influences topics obtained from transcriptions in Danish; a low resource language. First, we observe a baseline of cosine similarity scores between topic embeddings from automatic transcriptions and the descriptions of the podcasts written by the podcast creators. We then observe how the cosine similarities decrease as transcription noise increases and conclude that even when automatic speech recognition transcripts are erroneous, it is still possible to obtain high-quality topic embeddings from the transcriptions.