t0mm11y/voxscribe-live-caption

VoxScribe is a Windows desktop app that listens to audio, either sound coming out of your computer or a microphone, and turns it into live captions on screen. It works offline for the speech-to-text part, using a version of the Whisper model, and then translates that text in real time using Google Translate. The captions appear in a small, draggable window that always stays on top of other apps, so it is meant for things like meetings, presentations, or conversations between people who speak different languages. Right now it supports English, Simplified Mandarin, Traditional Mandarin, and Indonesian, both for the speech it listens to and the language it translates into. Converting between the two Mandarin scripts happens locally on your machine instead of going through an online translator. The app also includes a glossary that corrects technical or engineering terms after the speech recognition step, and it checks your computer's hardware to automatically pick a performance setting that fits. To avoid losing audio if your computer is too slow to keep up with transcription, VoxScribe temporarily saves incoming audio to disk instead of dropping it, and cleans that up automatically afterward. It also offers an optional local web address you can query to get the current caption, transcript, or app status, which could be useful if you wanted to build something else on top of it. Setup requires Windows 10 or 11, Python 3.10 or newer, at least 8GB of RAM, and roughly 6GB of free disk space for the speech model. You run it by cloning the code and launching a Python script, and there is also a way to package it into a single standalone Windows program. The project includes a written test suite and is released under the MIT license, which allows free use, including commercial use.