hytechnology26-prog/auto-srt-writer

This tool takes an English audio or video file and turns it into a pair of subtitle files, one in English and one in Simplified Mandarin, so you end up with bilingual subtitles ready to use. It works in two steps. First, it transcribes the spoken audio into an English subtitle file using a tool called stable-whisper, which is a version of OpenAI's Whisper speech recognition model that produces more accurate timing for each line. Second, it translates each subtitle line into Mandarin using a local AI model run through Ollama, a program that lets you run language models on your own computer instead of sending data to an outside service. The translation step is designed to handle short subtitle fragments carefully, using a glossary of pinned terms and the surrounding lines for context, so phrases that get split across lines are not translated in a way that loses their meaning. To use it, you need Python 3.10 or newer, a program called ffmpeg for reading audio and video files, and Ollama running locally with a translation model downloaded, such as the default qwen2.5:7b. Setup instructions are given separately for macOS, Linux, and Windows. If you have an Nvidia graphics card, the program will automatically use it to speed up transcription. Running the script on a video file produces two output files, one ending in _en.srt for English and one ending in _zh.srt for Mandarin. There are optional settings to choose a different output folder, use a larger and more accurate Whisper model, switch the translation model, add your own glossary of terms that should stay untranslated, skip transcription and only redo translation, or force the program to use a specific processor. The tool includes a built in glossary that keeps certain brand and product names untranslated, and if a line fails to translate for any reason, it falls back to keeping the original English text in the Mandarin file rather than leaving it blank.