collabora/WhisperLive ? reverse-engineered prompt

Build me a real time speech to text app that can transcribe both microphone input and audio files with very low delay, using Whisper under the hood.

I want a server and client setup where I can run the server locally, connect from a simple command line client, and see text appear as someone speaks. It should also support translating speech to English, speaker labels if possible, and word level timestamps for the transcript. Please make it work with different inference options so it can run on regular CPU setups and also support faster GPU friendly modes when available.

If it makes sense, include a simple web or browser based way to use it too, since I see browser extension folders in the project. Keep the setup as easy as possible, and add clear run commands and sensible defaults. Look up current docs online if you need to.