stt

Adding OpenAI text, speech, and Realtime paths without replacing the existing Gemini relay architecture

Stabilizing live interviews by defining reliable boundaries for user and assistant turns

Managing voice opt-in, waiting, playback, and cancellation as one product flow