What is whisper-real-time-transcription?

This is a fully offline, real-time speech transcription tool written in Rust, designed specifically to help people understand foreign-language audio as it plays. Everything runs locally on your computer, no data is sent to the cloud, no internet connection is needed. The app captures audio from your microphone or system audio (so you can transcribe a video call or media player), then runs it through a two-stage pipeline. First, a voice activity detector (a model that identifies when someone is actually speaking versus silence) segments the audio into phrases. Then two AI transcription models run in sequence: a fast, lightweight model gives you a rough preview almost instantly as you speak, and a slower but more accurate model processes the completed phrase and replaces the preview with clean final text. Short phrases skip the second pass entirely to save time. On top of transcription, the tool includes a word replacement dictionary so you can override specific transcribed words with your own translations, for example, automatically substituting a translated version of a place name whenever it appears. The interface is a live desktop window that shows transcription in real time. GPU acceleration is supported to speed up the AI models, with a fallback to CPU if no compatible GPU is available. You would use this if you are watching foreign-language video, listening to audio in a language you are learning, or doing any live monitoring of speech where you need quick, private, offline transcription. It is built in Rust and relies on the Whisper speech recognition model and Silero voice activity detection model running locally.