An Azure service that provides access to OpenAI’s GPT-3 models with enterprise capabilities.
Azure OpenAI Realtime SIP /refer endpoint returns 500 Internal Server Error
Hi,
we are currently testing the Azure OpenAI Realtime API via SIP and are running into a 500 error when trying to transfer an active SIP call using the documented /refer endpoint.
What works:
- Incoming SIP calls reach Azure OpenAI.
- We receive a realtime.call.incoming webhook.
- We use event.data.call_id from the webhook.
- The call can be accepted and becomes active.
What fails:
- Endpoint: https://<resource>.openai.azure.com/openai/v1/realtime/calls/<call_id>/refer
- When calling the documented Refer endpoint for the active call, Azure returns: HTTP 500 Internal Server Error Response body: Internal Server Error
- Example request: POST https://<resource>.openai.azure.com/openai/v1/realtime/calls/<call_id>/refer
- Headers: api-key: <Azure OpenAI API key> Content-Type: application/json
- Body: { "target_uri": "tel:+49157572*****" }
We also tested the request from Postman. The response is consistently 500 Internal Server Error. According to the documentation, the /refer endpoint should accept a target_uri such as +14155550123 or sip:agent@example.com and return 200 OK once the REFER is relayed to the SIP provider.
Can you give us a fix or work-around? Is this a service-side issue with the Realtime SIP /refer implementation or are there additional restrictions on target_uri formats, SIP URI transfers, or TLS transport parameters that are not documented? What can I do to redirect a call from the gpt-realtime model to a human?
Best regards, Elias!