Interactive voice chat using the SDK with streaming provider support (Gemini Live API).

Features

• Bidirectional Streaming: Real-time audio input/output through provider
• Pipeline Integration: Uses SDK pipeline with VAD middleware
• Provider-native TTS: Audio responses directly from provider
• Full-duplex Conversation: Seamless voice interaction

Requirements

• System:

  • Microphone (default system input)
  • Speakers/audio output
  • PortAudio library installed

• macOS:

  brew install portaudio

• Linux:

  sudo apt-get install portaudio19-dev

• Windows: Download and install PortAudio from http://www.portaudio.com/

• API Keys:

  • Gemini API key for streaming audio

Installation

cd sdk/examples/voice-chat
go mod tidy

Usage

1. Set your Gemini API key:

   export GEMINI_API_KEY=your-key-here

2. Run the example:

   go run .

3. Speak into your microphone

   • The session streams your audio to the provider
   • VAD middleware detects turn boundaries
   • Provider responds with text and/or audio
   • Audio responses play through speakers

4. Press Ctrl+C to exit

How It Works

This example uses the SDK’s proper pipeline architecture:

1. Provider Session: Creates streaming session with Gemini Live API
2. Bidirectional Session: Wraps provider session with SDK session management
3. Audio Capture: Microphone input sent as StreamChunks to session
4. Pipeline Processing: VAD middleware detects turns, provider generates responses
5. Response Handling: Text and audio responses received via response channel
6. Audio Playback: Provider-generated audio played through speakers

Architecture

┌─────────────┐
│ Microphone  │
└──────┬──────┘
       │ PCM chunks
       ▼
┌─────────────────────┐
│ BidirectionalSession│
│   (SDK Pipeline)    │
│                     │
│  ┌─────────────┐   │
│  │ VAD         │   │◄── Turn detection
│  │ Middleware  │   │
│  └─────────────┘   │
│         │           │
│         ▼           │
│  ┌─────────────┐   │
│  │ Provider    │   │◄── Gemini Live API
│  │ Session     │   │
│  └─────────────┘   │
│         │           │
└─────────┴───────────┘
          │
          ▼ Text + Audio
┌─────────────┐
│  Speakers   │
└─────────────┘

Customization

The pipeline handles VAD, transcription, and TTS internally through middleware. Configuration is done through the provider session request.

Next Steps

• Explore VAD demo for VAD configuration
• Check streaming example for text streaming
• See SDK documentation for pipeline middleware