Telugu Transcription for YouTube Videos Using LLMS

With the rapid increase in the amount of digital content, many people, especially non-English speakers, struggle to access information in their native language. This research presents a browser extension for automatic transcription of YouTube videos in Telugu using a fine-tuned version of the Open Al Whisper model. The fine-tuned model demonstrated a Word Error Rate (WER) of 15. 38% on the evaluation set, indicating a significant improvement over the base model. The tool provides real-time, accurate, and readable transcriptions directly on the video interface, helping Telugu speakers, including those with hearing impairments, understand video content easily. The system integrates with YouTube and processes audio using Flask in the backend. By optimizing the Whisper model specifically for Telugu, we enhance transcription quality and provide a seamless user experience. Furthermore, this tool can be beneficial for educational institutions, content creators, and language researchers who seek better accessibility for Telugu digital content. The implementation of this transcription system bridges the digital divide by making online video content more inclusive and user-friendly.