Whisperflu

Whisperflu is a lightweight, low-RAM voice-dictation desktop app for Windows. Speak naturally, and get clean, polished text inserted wherever your cursor is.

It was originally built with Electron and Node.js but has been fully rewritten in Flutter for maximum performance and minimal memory footprint.

🚀 Just want to use the app? (No setup required)

If you just want to use the app without messing with the code:

1. Download the latest v1.0.0 Release ZIP here!
2. Extract the ZIP file to any folder.
3. Open the whisperflu_dist folder and double-click whisperflu_app.exe to run the app.
4. An interactive Settings dialog can be opened by clicking the microphone icon. You can set up your local model or paste your Gemini API key there!

Features

• Press Left Alt to toggle recording.
• Uses a local whisper.cpp binary for fast, free speech-to-text.
• Cleans up transcripts using either the cloud-based Google Gemini API or a completely local, offline Qwen2.5-3B-Instruct model running via llama-server.exe.
• Auto-Download AI: If you toggle on Local AI but don't have the model downloaded yet, the app will securely download it in the background for you automatically!
• Personalized Lexicon (lexicon.json): Allows you to teach the AI specific acronyms, proper nouns, and unique vocabulary so it spells your words perfectly every time.
• Automatically pastes the cleaned text at your cursor's current position.
• Frameless, transparent floating UI overlay on the bottom right of your screen.
• Drag to move the window, resize from the bottom-right corner.

Tech Stack

• Flutter (Windows Desktop app)
• SoX (Audio recording)
• whisper.cpp (Local STT)
• llama.cpp (Local LLM server)
• Gemini API (Cloud cleanup fallback)

🛠️ For Developers (Making Changes)

If you want to modify the source code:

1. Install Flutter: Ensure you have Flutter installed and configured for Windows desktop development (requires Visual Studio build tools).
2. The Flutter source code is located in the whisperflu_app/ directory.
3. Open a terminal in whisperflu_app/ and run:

   flutter pub get
   flutter run -d windows

4. Building for Production:

   flutter build windows

5. After building, you can copy the contents of whisperflu_app/build/windows/x64/runner/Release/ into the whisperflu_dist/ folder to distribute your updated version alongside the whisper.cpp, llama.cpp, and sox binaries.

Usage

1. Keep the app running in the background.
2. Focus on any text field in any application (e.g., Word, Notion, browser).
3. Press F8 to start recording. The floating UI will turn red and show "RECORDING".
4. Speak naturally.
5. Press F8 again to stop. The indicator will show "TRANSCRIBING" then "CLEANING".
6. The cleaned text will be automatically pasted into your application.