Improving transformer performance for text summarization in video transcription

Putra, Rizky Dwi and Atmadja, Aldy Rialdy and Gerhana, Yana Aditia (2025) Improving transformer performance for text summarization in video transcription. Journal of Computer Engineering, Electronics and Information Technology (COELITE, 4 (2). pp. 1-8. ISSN 2829-4149

Preview

Text Improving Transformer Performance for Text Summarization in Video Transcription.pdf Download (401kB) | Preview

Abstract

In conveying information today, it can take the form of online video content. The rapid growth of online video content has created a strong need for automatic text summarization to improve information efficiency. Summarization is important because it allows audiences to quickly capture the essence of lengthy materials, reduces information overload, and ensures that key points can be accessed without going through the entire content. This study explores the use of Whisper Turbo for transcription and mT5 for summarizing Indonesian-language YouTubevideos. Whisper Turbo produces accurate transcriptions, although the results vary depending on audio quality and topic complexity. The transcribed text is then summarized using mT5, which achieves strong performance with a ROUGE-1 F1 score of 54.13% and aROUGE-L score of 49.39%. These findings indicate that mT5 outperforms the standard T5 model despite using less training data. Overall, the combination of Whisper Turbo and mT5 offers an effective solution for generating concise and reliable summaries of video content, with broad potential applications in education, journalism, and digital documentation.

Actions (login required)

View Item