ProveWorksSign in
← All architecture templates
10VoiceBlueprint

Native realtime speech-to-speech

A realtime multimodal model consumes audio and emits audio without a separately visible STT/TTS cascade.

Runtime flow

What happens, in order.

The flow becomes the backbone of the audit trace. Every important boundary needs an input, an output, timing, and an owner.

  1. 01

    Authorize and configure the realtime session

  2. 02

    Stream audio and turn-detection events

  3. 03

    Execute permitted tools with correlated events

  4. 04

    Stream audio output and preserve a normalized trace

Components

System inventory

  • Session broker
  • Realtime model
  • VAD
  • Tool bridge
  • Audio client

Test packs

Required coverage

  • Session security
  • Turn taking
  • Audio latency
  • Tool correlation
  • Safety

Evidence

Capture contract

  • Session config
  • Audio events
  • Transcripts
  • Tool events
  • Latency markers