Scholarlink

This chapter examines the quality of automatically generated subtitles produced by two different machine translation systems: YouTube, which uses a classical pipeline model, and OpenAI’s end-to-end model Whisper. The aim of this study is to evaluate the performance of these systems using six popular science YouTube videos. The quality assessment is based on the FAR model presented by Jan Pedersen (2017). The results of the sample-based and therefore only partially representative study show that Whisper seems to perform significantly better overall than YouTube; the difference in the area of readability is particularly striking. The chapter also addresses some limitations of the FAR model when analyzing automatically generated subtitles.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Maschinelle Untertitelung mittels Pipeline- und End-to-End-Ansatz. Ein Vergleich von Youtube und Whisper

  • 章节

  • 发表日期：

    2025年

查看全文
  • Lea Hof

摘要

This chapter examines the quality of automatically generated subtitles produced by two different machine translation systems: YouTube, which uses a classical pipeline model, and OpenAI’s end-to-end model Whisper. The aim of this study is to evaluate the performance of these systems using six popular science YouTube videos. The quality assessment is based on the FAR model presented by Jan Pedersen (2017). The results of the sample-based and therefore only partially representative study show that Whisper seems to perform significantly better overall than YouTube; the difference in the area of readability is particularly striking. The chapter also addresses some limitations of the FAR model when analyzing automatically generated subtitles.