Flototext

Windows voice recognition application with real-time transcription.

Features

• Push-to-talk with F2: Hold F2 to record, release to transcribe
• AI Transcription: Uses Qwen3-ASR-1.7B (supports French and 52 languages)
• Auto-paste: Transcribed text is automatically pasted at cursor position
• Custom Dictionary: Define word corrections for technical terms, names, etc.
• Auto-mute: System audio is muted during recording to prevent interference
• History: All transcriptions are saved (7-day retention)
• Visual feedback: Color-coded system tray icon + Windows notifications

Requirements

• Windows 10/11
• Python 3.10+
• NVIDIA GPU with CUDA (RTX series recommended)
• ~4 GB VRAM available

Installation

1. Clone the project:

git clone https://github.com/florentcollect/Flototext.git
cd Flototext

2. Create a virtual environment (recommended):

python -m venv venv
venv\Scripts\activate

3. Install dependencies:

pip install -r requirements.txt

4. For GPU support with PyTorch CUDA:

pip install torch torchaudio --index-url https://download.pytorch.org/whl/cu121

5. Double-click install.bat (as administrator)
   • Configures auto-start with Windows
   • Registers the application in Windows Settings

To uninstall: Windows Settings → Apps → Flototext → Uninstall

Usage

1. Launch manually (if needed):

   • Double-click start-flototext.bat
   • Or: python -m flototext.main

2. The icon appears in the system tray (near the clock)

3. Recording:

   • Press and hold F2 to start recording
   • Speak (supports 52 languages including French, English, Chinese...)
   • Release F2 to stop and transcribe

4. The transcribed text will be automatically pasted at your cursor position

Tray Icon States

Color	State
Orange	Loading model
Green	Ready
Red	Recording
Yellow	Processing transcription
Gray	Error

System Tray Menu

Right-click on the icon to access options:

• Copy last transcription: Copy the last transcription to clipboard
• Edit dictionary: Open the custom words dictionary file
• Sounds: Enable/disable audio feedback
• Notifications: Enable/disable Windows notifications
• Mute during recording: Enable/disable system audio muting while recording
• Quit: Close the application

Custom Dictionary

Create custom word corrections for terms that are frequently misrecognized (technical jargon, names, etc.).

The dictionary file is located at data/custom_words.json:

{
  "corrections": {
    "clode": "Claude",
    "anthropique": "Anthropic",
    "pie torche": "PyTorch"
  }
}

• Keys: What the ASR model outputs (lowercase)
• Values: The correct spelling you want

Access via tray menu: Edit dictionary (opens the file in your default JSON editor)

Database

Transcriptions are stored in data/transcriptions.db and kept for 7 days.

View history:

sqlite3 data/transcriptions.db "SELECT * FROM transcriptions ORDER BY created_at DESC LIMIT 10"

Project Structure

Flototext/
├── flototext/
│   ├── __init__.py
│   ├── main.py                 # Entry point
│   ├── config.py               # Configuration
│   ├── core/
│   │   ├── hotkey_manager.py   # F2 key detection
│   │   ├── audio_recorder.py   # Microphone capture
│   │   ├── transcriber.py      # Qwen3-ASR model
│   │   ├── text_inserter.py    # Clipboard paste
│   │   ├── text_corrector.py   # Custom word corrections
│   │   └── audio_muter.py      # System audio muting
│   ├── storage/
│   │   ├── database.py         # SQLite operations
│   │   └── models.py           # Data models
│   └── ui/
│       ├── tray_app.py         # System tray icon
│       ├── notifications.py    # Toast notifications
│       └── sounds.py           # Audio feedback
├── data/
│   ├── transcriptions.db       # Database
│   └── custom_words.json       # Custom word dictionary
├── assets/
│   └── icon.ico
├── install.bat                 # Windows installer
├── uninstall.bat               # Uninstaller
├── start-flototext.bat                   # Manual launch
├── requirements.txt
└── README.md

Configuration

Edit flototext/config.py to customize:

• hotkey.trigger_key: Trigger key (default: "f2")
• audio.sample_rate: Sample rate (default: 16000)
• model.model_name: ASR model to use
• ui.play_sounds: Sounds enabled by default
• ui.show_notifications: Notifications enabled by default
• ui.mute_during_recording: Mute system audio while recording (default: True)

Troubleshooting

Model won't load

• Check CUDA is installed: python -c "import torch; print(torch.cuda.is_available())"
• Make sure you have enough VRAM (~4 GB)

No audio during recording

• Check that the default microphone is properly configured in Windows
• Test with another recording application

Text doesn't paste

• Make sure a text field is active (blinking cursor)
• Text is also copied to clipboard (use Ctrl+V manually)

Support

If Flototext saves you time, consider supporting its development:

License

MIT License