AI Window Assistant (Gemini Live Edition)

AI Window is a minimalist, transparent window assistant that combines real-time voice interaction with immersive background scenes. Powered by Gemini Live and YouTube, it allows you to transform your workspace with just a voice command.

🌟 Features

• Real-time Voice Interaction: Bidirectional streaming using gemini-2.5-flash-native-audio-preview.
• Smart Scene Switching: Automatically searches and plays YouTube 4K window views (e.g., "Rainy London", "Swiss Alps") based on conversation keywords.
• Intelligent Audio Management:
  • Auto-Pause: Background music/video automatically pauses when you start talking to the AI.
  • Auto-Resume: Background audio resumes seamlessly once the conversation ends.
  • Jitter Buffer: Advanced latency management to prevent audio stuttering during network bursts.
• Minimalist UI: A sleek, transparent, and "always-on-top" PyQt6 interface.
• Auto Mic Closure: The assistant automatically closes the microphone 6 seconds after a search command is detected, allowing it to finish its verbal confirmation.

🛠️ Prerequisites

• Python 3.10+
• FFmpeg: Required for audio processing.
• MPV: Required for background video playback.
• yt-dlp: Required for searching YouTube content.
• Dependencies:

  pip install PyQt6 google-genai nest_asyncio

🚀 Setup & Launch

1. Get a Gemini API Key: Visit the Google AI Studio to get your key.
2. Set Environment Variable:

   export GEMINI_API_KEY='your_api_key_here'

3. Launch the Application: Everything is automated via the start script. Simply run:

   chmod +x start_window.sh
   ./start_window.sh

   This script will automatically clear previous sockets, start MPV in the background with a default rainy scene, launch the AI UI, and clean up processes upon exit.

🎙️ Usage

• Voice Command: Click the 🎤 button to start a Live session.
• Switch Scenes: Tell the AI something like:
  • "I want to see the rainy streets of London."
  • "Show me a snowy mountain view."
  • "帮我换成日本街道的风景" (Support for Traditional Chinese).
• Text Entry: You can also type commands into the input field at the bottom.
• Exit: Click the '✕' or press Esc.

⚙️ Technical Details

• Audio Configuration:
  • Recording: 16kHz, 16-bit PCM.
  • Playback: 24kHz, 16-bit PCM (standard for Gemini Live output).
• Jitter Buffer: Built with a 5-second burst tolerance and 20ms check intervals to ensure smooth playback regardless of network conditions.
• Device Selection: Automatically prioritizes external microphones (USB Audio, ConferenceCam) for better voice quality.

📝 License

This project is for demonstration and personal use. Powered by Google Gemini.