How to Deploy Voice Live API Models

Question

Hi Everyone,

I have a doubt!

How can we deploy the Voice Live API Models (E.g. Customer Service Feature in AI Foundry Playground) to production, so that everyone can use.

I want to deploy it for production!

Thanks in Advance!

Answer 1

Hi Ashwath Bala S,

To deploy the Voice Live API Models (e.g. Customer Service feature in AI Foundry Playground) to production, follow the steps below:

Step-by-Step Production Deployment of Voice Live API Models:

Step1: Provision Voice Live Model in Azure Foundry

1.   Go to: https://ai.azure.com/foundry

2.   Choose the Customer Service preset or Voice Live (Realtime) model.

3.   Click Deploy.

4.   After deployment, note the:

o   Endpoint URL (e.g., wss://.../realtime)

o   API key or Entra ID for authentication.

Step2: Build a Client App (Web or Mobile)

Use WebSocket or WebRTC to stream voice to and from the deployed model.

·   Azure provides sample SDKs in:

o   Python (WebSocket-based)

o   Node.js (WebRTC-based)

Example (Python WebSocket client):

import websockets
import asyncio
async def voice_chat():
    uri = "wss://your-endpoint.realtime.azure.com"
    async with websockets.connect(uri, extra_headers={"Authorization": "Bearer YOUR_KEY"}) as ws:
        await ws.send(your_audio_bytes)  # Send microphone input
        response = await ws.recv()       # Receive spoken response
        print(response)
asyncio.run(voice_chat())

Step3: Integrate Frontend & Backend

·   Frontend: Use JavaScript to capture mic input and play audio responses.

·   Backend (Optional): Proxy requests, manage authentication, rate limiting, etc.

Step4: Deploy to Production

Choose one of the following Azure services to host your app:

·   Azure App Service – For websites and APIs

·   Azure Container Apps – For containerized apps

·   Azure Static Web Apps + Azure Functions – For serverless deployment

Step5: Make It Publicly Usable

·   Secure the API with Azure API Management or a backend token exchange.

·   Add custom domain + HTTPS.

·   Optional: Add usage tracking, authentication (Azure AD B2C), and failover.

Please refer below documents for more information:

How to use the GPT-4o Realtime API for speech and audio (Preview)

GPT-4o Realtime API for speech and audio (Preview)

And also sample code repo form git:

https://github.com/Azure-Samples/aoai-realtime-audio-sdk

Hope this helps. If you have any follow-up questions, please let me know. I would be happy to help.

**

Please do not forget to "Accept the answer” and “up-vote” wherever the information provided helps you, this can be beneficial to other community members.

Thank you! 

Answer 2

Hi Ashwath,

Thank you for your contacting Q&A Forum.

The Voice Live API Models feature is currently in preview and not suitable for production deployment. At the moment, only code example quickstarts are available for testing and development purposes.

You can find the quickstart guide and code examples here: https://learn.microsoft.com/en-us/azure/ai-services/speech-service/voice-live-quickstart?tabs=windows%2Capi-key

If I have answered your question, please accept this answer as a token of appreciation and don't forget to give a thumbs up for "Was it helpful"!

Best Regards,