Skip to content

AuDesign Voice — Documentation

Web demo

  • Listening
  • User speaking
  • Agent speaking

A self-hosted, UAE-resident alternative to Azure's Voice Live API. Plug it into any LLM-based application via WebSocket or via the official SDKs.

UAE North architecture

Read in order

  1. Getting started — install, run, and have a voice conversation in 5 minutes.
  2. Wire protocol — every WebSocket event, audio format, and state transition.
  3. LLM backends — wire the service to Azure OpenAI, OpenAI, or a Microsoft Foundry project.
  4. Microsoft Agent Framework integration — expose Agent Framework agents through AuDesign Voice.
  5. SDKs — Python and TypeScript packages: install, publish, embed.
  6. Deployment — Bicep + App Service for UAE North.
  7. Operations & residency — quotas, logs, the residency caveat.

How a turn flows

Conversation sequence

Session state machine

Session state machine

What you get

Capability Voice Live (Azure) AuDesign Voice
UAE North data plane ❌ not available
OpenAI-Realtime-style WS protocol
Barge-in, server VAD, end-of-turn
Tool / function calling passthrough
Arabic (ar-AE) + English code-switch
RAG over your knowledge base ✅ (VoiceRAG) ✅ (Azure AI Search)
Custom Neural Voice
Bring your own LLM (OpenAI / Foundry)
Full source control
Avatar (TTS Avatar) ❌ (UAE region)
Native realtime audio model ❌ (UAE region)

SDKs

Language Package Install
Python (≥ 3.10) async audvoice-client pip install audvoice-client
TypeScript / browser @audvoice/client npm i @audvoice/client
Any other language Talk to the WebSocket directly See protocol