sihyunadventure/dubyduby

dubyduby is a command line tool that takes any YouTube video URL and produces a Korean dubbed version of that video, complete with synchronized Korean subtitles burned into the output file. The tool works in three phases. First, it automatically downloads the video using yt-dlp and transcribes the audio using Soniox, a speech to text service that also separates speakers. This produces a timestamped transcript. Second, and this is what makes the tool different from other dubbing tools, rather than calling a fixed AI model, the user's own AI agent, whichever assistant they already use day to day such as Claude Code or Codex, reads the transcript and writes a structured JSON file with English and Korean sentence pairs, following a style guide provided by the project. No separate AI API key is needed for this translation step. Third, running the same command again detects the completed translation and continues automatically. It adjusts the timing of each sentence to fit within its original slot, speeding speech up slightly if needed, synthesizes Korean audio using Supertonic voices, and burns in subtitles before packaging everything into a final mp4 file. Supertonic provides ten Korean voices to choose from, five male labeled M1 through M5 and five female labeled F1 through F5, with automatic speaker assignment based on pitch analysis for videos with multiple speakers. A glossary file handles cases where the speech to text step mishears proper nouns. The project runs on Python and requires a Soniox account for transcription, which includes 200 free minutes per month, plus ffmpeg. No GPU is needed and no LLM API key is needed. The README is written in Korean.