Voice Flow V2

A lightning-fast, ultra-accurate voice dictation and transcription tool for Windows. Hold Ctrl+Alt (or tap to toggle), speak naturally, and your voice is transcribed with near-instant speed and pasted directly into whatever app has focus.

Powered by faster-whisper with NVIDIA CUDA GPU acceleration, Smart Dictation post-processing, and multi-engine cloud support (OpenAI, Groq, and Gemini).

---

What's New in V2

• ⚡ CUDA GPU Acceleration: Automatically detects your NVIDIA GPU (e.g. RTX 4050) to run Whisper in ~200–300ms with float16 precision (gracefully falls back to CPU if no GPU is present).
• 🎯 State-of-the-Art Accuracy (large-v3-turbo): Uses Whisper's 809M-parameter turbo model by default on GPU for ~99% accuracy across accents, terminology, and complex phrasing.
• 🧠 Smart Dictation Engine:
  • Spoken Punctuation: Naturally handles voice commands like "period", "comma", "question mark", "new line", "colon", "open quotes", etc.
  • Verbal Filler Cleanup: Filters stuttered duplicates and filler words ("um", "uh", "ah").
  • Anti-Hallucination Guard: Eliminates phantom Whisper phrases ("Thank you for watching", "Subtitles by...", music tags) during silence or breathing.
  • AI Polish (Optional): Can route transcripts through an LLM to rewrite informal rambling into structured, publication-ready prose.
• 🎙️ Live Audio Level VU Meter: Real-time visual feedback in the floating widget during recording so you can verify microphone activity.
• 🎛️ Dual Trigger Styles: Choose between Push-to-Talk (Hold) or Tap-to-Toggle (Hands-Free) in Settings.
• ⚙️ Graphical Settings Window: Configure everything directly from the UI:
  • Select your preferred microphone from a dropdown with a built-in Live Mic Test meter.
  • Choose between Local Whisper models (large-v3-turbo, medium, small, base, tiny) or Cloud engines (OpenAI, Groq, Gemini).
  • Input API keys securely in the GUI.
  • Customize hotkeys and trigger modes.
• 📋 Transcript History with One-Click Copy: Copy any past transcript with a single click, or clear history when done.
• 🔔 Subtle Sound Cues: Gentle audio chimes when recording starts and finishes.
• 💾 Persistent Settings: Configurations are saved to %LOCALAPPDATA%\VoiceFlow\config.json.

---

Quick Start (Run Executable)

Double-click the Voice Flow shortcut on your Desktop, or run dist\VoiceFlow.exe. No Python installation or terminal needed.

• To talk: Hold Ctrl+Alt (or your custom hotkey), speak, then release.
• To configure: Click the ⚙ Settings button in the floating widget (or right-click the system tray icon and select Settings...).
• To switch mode: Click the Local or Cloud pill on the widget.
• To quit: Click Quit on the widget or right-click the tray icon.

---

Developer Setup (Running from Source)

# Create and activate virtual environment
python -m venv venv
venv\Scripts\activate

# Install dependencies
pip install -r requirements.txt

# Run the app
python main.py

Rebuilding the .exe

venv\Scripts\activate
pyinstaller --onefile --windowed --name VoiceFlow --icon icon.ico --collect-data faster_whisper --collect-binaries ctranslate2 --distpath dist --workpath build --specpath build main.py

This regenerates dist\VoiceFlow.exe.

---

Cloud Providers (Optional)

If you prefer online cloud transcription or want to use AI Polish:

1. Open Settings (⚙ icon on widget).
2. Under Engine & Model, select your cloud provider:
   • OpenAI: Uses whisper-1 for transcription.
   • Groq: Uses ultra-fast whisper-large-v3 (~150ms cloud latency).
   • Google Gemini: Uses gemini-2.5-flash for multimodal speech-to-text.
3. Paste your API key and click Save & Apply.

---

Configuration & Logs

• Config File: %LOCALAPPDATA%\VoiceFlow\config.json
• Application Log: %LOCALAPPDATA%\VoiceFlow\voiceflow.log