ggml-org/whisper.cpp ? reverse-engineered prompt

Build me a fast local speech to text tool using OpenAI Whisper that runs fully offline. I want to be able to download a model, build the app, and run one simple command to transcribe an audio file into text.

Please make it lightweight and portable, focused on C or C++, with a command line version first. It should work on normal laptops, especially Mac, Linux, and Windows, and it should be able to run on CPU even if there is no fancy GPU. If the machine supports Apple Silicon or other acceleration, use it when available, but don’t make that required.

Include a sample audio file, a quick demo command, and clear setup steps so I can test it right away. Handle the basic audio format issue too, so if the input is not a 16 bit WAV, explain or add a simple conversion step with ffmpeg. Look up the current docs online if you need to.