kizuna-ai-lab/sokuji v0.15.28

on GitHub

What's New

Gemini Voice Activity Detection (VAD) Configuration

• Added VAD settings panel for Gemini provider with configurable parameters:
  • Start / End of Speech Sensitivity (High / Low)
  • Silence Duration (50ms – 3000ms)
  • Prefix Padding (0ms – 2000ms)
• Added Push-to-Talk mode for Gemini — manually control turn boundaries using the Space key or mic button
• Changed default activityHandling from NO_INTERRUPTION to START_OF_ACTIVITY_INTERRUPTS for better turn splitting
• Participant audio client always uses Auto mode (no PTT)

Microsoft Teams Support

• Added support for the new teams.cloud.microsoft domain in the browser extension

Translations

• Added Gemini VAD setting translations for all 30 supported languages

Known Issues

• silenceDurationMs parameter has no effect on gemini-3.1-flash-live-preview — it works correctly on gemini-2.5-flash models. Details & reproduction

Full Changelog: v0.15.27...v0.15.28