what is wisprgemma fr?

WisprGemma is a voice dictation tool that turns spoken words into clean, written text, and it does all of that work directly inside your browser instead of sending your voice to a company's server. It comes in two forms: a web app you open in a browser tab, and a Chrome extension that lets you dictate directly into any text field on any webpage. The project uses a single AI model, Gemma 4 E2B, to handle everything in one step: recognizing your speech, cleaning it up, and rewriting it if needed. There is no separate speech recognition tool and no server involved, since the model runs on your own computer's graphics card through a browser technology called WebGPU. The first time you use it, the app downloads about 3.5 gigabytes of model files, but after that it works offline. Before speaking, you choose one of four modes. Clean dictation removes filler words and applies spoken commands like starting a new paragraph. Verbatim mode transcribes exactly what you said with no cleanup. Polished email mode rewrites what you say into a professional sounding email. The fourth mode transcribes speech in any language and outputs the result in English. The web app tracks statistics like words dictated, typing speed saved, and a streak chart, stored only in your browser. The Chrome extension adds a side panel with push to talk dictation on any page, and includes workarounds for Chrome's extension permission system, since Chrome cannot ask for microphone access directly inside a side panel. To try the web app, a developer runs a local server command and opens it in Chrome or Edge version 121 or newer with WebGPU turned on. The Chrome extension is loaded manually as an unpacked extension through Chrome's developer mode. Both pull the AI model from Hugging Face, a site that hosts AI models, and cache it locally afterward. The README does not state a license.