What is yt-realtime-translate?

This project is a Chrome browser extension that captures the audio playing in an open YouTube tab and transcribes what is being said in real time, showing the text in a side panel next to the video. It is described as an MVP, meaning a minimal early version of the idea rather than a finished polished product. Instead of using the microphone or a cloud speech service, the extension grabs the tab's own audio directly through Chrome's tab capture feature, so it does not need microphone permission and no audio ever leaves your computer. That audio is processed through a local speech recognition server that you run yourself on your machine, built with Python's FastAPI framework and a project called faster-whisper, which is a fast version of OpenAI's Whisper speech recognition model. The extension streams audio to this local server over a WebSocket connection, gets back partial and finalized text as people speak, and displays it live in a React-based side panel rather than as subtitles overlaid on the YouTube video itself. If an AI translation service is configured in the extension's settings, the finalized transcript lines can also be translated shortly after they appear. To use it, you need Node.js, a recent version of Chrome, and Python 3.9 or newer for the local Whisper server. You build the extension with npm, load it into Chrome as an unpacked extension from the generated output folder, then start the local Whisper server separately before opening a YouTube tab and clicking the extension icon. If the local server is not running, the panel simply reports that it is disconnected while audio playback continues normally either way. The project is structured so the underlying speech recognition engine can be swapped out, with an older experimental implementation based on a different toolkit called Vosk still present in the code but not currently used. The README does not state a license for the project.