From Podcast to Podcast: Automated Content Localization Using OpenAI API Stack

Abstract

Large Language Models have revolutionized content localization, enabling complex transformations while preserving nuanced context and style. In this talk, we’ll explore a real-world implementation of an automated podcast localization system that leverages multiple LLM capabilities. Through a Kotlin-based solution, we’ll demonstrate how different LLM strengths complement each other: Whisper for accurate speech recognition, GPT-4 for context-aware text refinement and translation, and an LLM-powered TTS for natural voice synthesis. We’ll dive deep into prompt engineering practices that enable accurate transcription cleanup, style preservation during translation, and handling of mixed-language content. The session will highlight practical aspects of building production-ready LLM pipelines, including chunking strategies for token limitations and maintaining content authenticity across language barriers.