cl1107/video-translate

Video Translate Assistant is a desktop application, built with Electron, that turns videos into translated subtitles or into readable Markdown notes, and it does almost all of its work on your own computer instead of sending files to a cloud service. Its README is written in Chinese and targets users who need to localize or transcribe video content. The app offers two separate workflows. The subtitle workflow takes a local video file or an online link from a site like YouTube or Bilibili, downloads it if needed, and either reuses subtitles already provided by the platform or generates them with local speech recognition when none exist. Recognized or platform provided text can then be optionally polished and translated into another language using a locally running Ollama language model, and exported as SRT or ASS subtitle files, with an option to burn the subtitles directly into the video. The second workflow, the transcript pipeline, takes the same kind of source and turns an entire video into a structured Markdown document that can be previewed full screen or exported, useful for meeting notes or course recordings. Speech recognition runs locally through sherpa-onnx, with a small model called SenseVoice used by default for Chinese, English, Japanese, Korean, and Cantonese audio, downloaded automatically the first time it is needed. A larger alternative model is available for noisier or dialect heavy audio but must be downloaded manually. The interface is built with React, TailwindCSS, and Base UI, and supports a dark mode. Task progress, pausing, resuming, and processing logs are all tracked in a local SQLite database. To run it, you need Node.js 22.13 or newer, at least 8GB of memory, and about 10GB of free disk space for the models. Setup also requires installing FFmpeg for audio and video handling and Ollama for local translation, then installing the project's dependencies with pnpm before starting the desktop app in development or production mode. Prebuilt release packages come in two variants, one bundling FFmpeg and one relying on a system installed copy. The project is released under the MIT license.