Browse documentation

Voice Architecture & Acoustic Intelligence

Zaya utilizes bidirectional full-duplex WebSocket audio streaming powered by Google Gemini Live (models/gemini-3.8-live).


Core Audio Capabilities

1. Acoustic Barge-In (Interruption)

You can interrupt Zaya at any moment while she speaks. An on-device WebRTC VAD filter continuously samples your microphone. When your voice begins, the desktop speaker buffer is immediately truncated, ducked audio restores, and active listening resumes.

2. Windows CoreAudio Ducking

During spoken dialogue (either yours or Zaya's), background computer audio drops to 20% of its previous volume. When the turn concludes, background sound smoothly fades back to 100%.

"audio_ducking": {
  "enabled": true,
  "background_ratio": 0.2,
  "voice_volume": 1.0,
  "release_seconds": 0.8
}

3. Nika Voice Persona

Zaya's voice synthesis is tuned to the Nika personality delivery model, expressing nuanced affective delivery based on conversation context.

Something unclear or not working?

Troubleshooting guide ↗Report a problem ↗