Audio Streaming

Audio Input (WebSocket)
    ↓
Continuous Stream
    ↓
VAD + STT (parallel)
    ↓
Turn Detection
    ↓
Agent Response

┌─────────────────────────────────────┐
│ Audio Input (WebSocket)             │
│ - client_audio_start               │
│ - client_audio_chunk (continuous)   │
│ - client_audio_end                  │
└──────────────┬──────────────────────┘
               │
               v
┌─────────────────────────────────────┐
│ VAD (Voice Activity Detection)      │
│ - START_OF_SPEECH                   │
│ - END_OF_SPEECH                     │
│ - CONTINUING                        │
└──────────────┬──────────────────────┘
               │
               v (parallel processing)
┌─────────────────────────────────────┐
│ STT (Speech-to-Text)                │
│ - Streams audio continuously         │
│ - Provides interim transcripts       │
│ - Provides final transcripts        │
└──────────────┬──────────────────────┘
               │
               v
┌─────────────────────────────────────┐
│ Turn Detection                      │
│ - Analyzes conversation context     │
│ - Predicts end-of-turn probability  │
│ - Applies dynamic endpointing       │
└──────────────┬──────────────────────┘
               │
               v
┌─────────────────────────────────────┐
│ AI Agent (LLM)                      │
│ - Processes transcript               │
│ - Generates response                │
└─────────────────────────────────────┘