gpt-realtime 2.0 on Azure OpenAI stops emitting audio mid-response (audio output truncates); identical setup works on OpenAI's direct API

Question

gpt-realtime 2.0 on Azure OpenAI stops emitting audio mid-response (audio output truncates); identical setup works on OpenAI's direct API

Abdul Rehman 65

When using the gpt-realtime 2.0 model deployed via Azure OpenAI, the model intermittently stops producing audio output mid-session — it behaves as though it simply stops talking. The same application code, prompts, and audio configuration work correctly against OpenAI's direct Realtime API; the failure only occurs against the Azure-hosted deployment. This points to a difference in the Azure hosting/serving layer rather than in our client or in the model itself.

Expected behavior:
The model streams a complete audio response for each turn (response.audio.delta events through to response.audio.done / response.done with status: "completed"), matching the behavior we observe on OpenAI's direct API.

Actual behavior:
Mid-session, audio output stops. The model "goes silent" and no further audio is produced for that turn (or the turn ends prematurely). This recurs multiple times within a single session. See the attached recording — there are repeated silent stretches of roughly 1.5–3.8 seconds where expected audio is missing.

Environment

Model / deployment: gpt-realtime 2.0 (Azure deployment name: [OnlimInternalTesting-production-manual])
Azure OpenAI resource region: [francecentral]
Audio format: PCM16, 24 kHz, mono (input and output)
Comparison baseline: OpenAI direct Realtime API, same model family, same client code

SRILAKSHMI C 19,550 Reputation points Microsoft External Staff Moderator

2026-06-30T12:09:12.38+00:00
Hello @Abdul Rehman

Thank you for Reaching out to Microsoft Q&A.

From your explanation, I understand that gpt-realtime 2.0 deployed through Azure OpenAI intermittently stops streaming audio in the middle of a response. The same client application, prompts, audio format (PCM16, 24 kHz, mono), and session configuration work as expected when connected to the OpenAI direct Realtime API, where the response continues until the expected response.audio.done and response.done events are received.

Since the issue occurs only with the Azure-hosted deployment and not with the OpenAI direct API, it suggests that the behavior may be related to the Azure OpenAI service or deployment rather than the client implementatio

Before considering a service-side issue, we recommend verifying the following:

Ensure you're using the latest supported Azure OpenAI Realtime API endpoint and API version for gpt-realtime.

Confirm that your client continues processing all streamed WebSocket events until both:

response.audio.done

response.done (with status: "completed") are received.

Review the session logs to determine whether the WebSocket connection remains active when the audio stops or whether any error, interruption, or close event is generated.

Since you've confirmed the same application works against the OpenAI direct API, it would be helpful to compare:

The sequence of Realtime events received from Azure versus OpenAI.

Whether Azure stops emitting response.audio.delta events before sending response.audio.done.

Whether the session terminates normally or remains open after the audio stops.

Thank you!
SRILAKSHMI C 19,550 Reputation points Microsoft External Staff Moderator

2026-07-01T11:34:25.5933333+00:00

Hi @Abdul Rehman

Did you get any chance to review the above response. Do let me know if you have any further queries.

Thank you!
SRILAKSHMI C 19,550 Reputation points Microsoft External Staff Moderator

2026-07-02T01:59:45.34+00:00

Hi @Abdul Rehman

We haven’t heard from you on the last response and was just checking back to see if you have a resolution yet. In case if you have any resolution please do share that same with the community as it can be helpful to others. Otherwise, will respond with more details and we will try to help.

Thank you!

1 answer

Your answer

SRILAKSHMI C 19,550 Reputation points Microsoft External Staff Moderator

2026-06-30T12:09:12.38+00:00

Hello @Abdul Rehman

Thank you for Reaching out to Microsoft Q&A.

From your explanation, I understand that gpt-realtime 2.0 deployed through Azure OpenAI intermittently stops streaming audio in the middle of a response. The same client application, prompts, audio format (PCM16, 24 kHz, mono), and session configuration work as expected when connected to the OpenAI direct Realtime API, where the response continues until the expected response.audio.done and response.done events are received.

Since the issue occurs only with the Azure-hosted deployment and not with the OpenAI direct API, it suggests that the behavior may be related to the Azure OpenAI service or deployment rather than the client implementatio

Before considering a service-side issue, we recommend verifying the following:

Ensure you're using the latest supported Azure OpenAI Realtime API endpoint and API version for gpt-realtime.

Confirm that your client continues processing all streamed WebSocket events until both:

response.audio.done

response.done (with status: "completed") are received.

Review the session logs to determine whether the WebSocket connection remains active when the audio stops or whether any error, interruption, or close event is generated.

Since you've confirmed the same application works against the OpenAI direct API, it would be helpful to compare:

The sequence of Realtime events received from Azure versus OpenAI.

Whether Azure stops emitting response.audio.delta events before sending response.audio.done.

Whether the session terminates normally or remains open after the audio stops.

Thank you!
SRILAKSHMI C 19,550 Reputation points Microsoft External Staff Moderator

2026-07-01T11:34:25.5933333+00:00

Hi @Abdul Rehman

Did you get any chance to review the above response. Do let me know if you have any further queries.

Thank you!
SRILAKSHMI C 19,550 Reputation points Microsoft External Staff Moderator

2026-07-02T01:59:45.34+00:00

Hi @Abdul Rehman

We haven’t heard from you on the last response and was just checking back to see if you have a resolution yet. In case if you have any resolution please do share that same with the community as it can be helpful to others. Otherwise, will respond with more details and we will try to help.

Thank you!

Answer 1

Hello Abdul Rehman,

Welcome to the Microsoft Q&A and thank you for posting your questions here.

I understand that gpt-realtime-2 on Azure OpenAI is intermittently stopping audio output mid-response, while the same application, prompt, and PCM16 24 kHz mono audio setup works correctly against OpenAI’s direct Realtime API.

In addition to @SRILAKSHMI C

The most reliable fix is to run the Azure Realtime integration on the GA /openai/v1 Realtime endpoint, pass the Azure deployment name as the model value, and use WebRTC for live user-facing audio. Then reproduce the issue while logging all Realtime events. If Azure stops emitting response.audio.delta and either sends response.done as completed with missing audio or never sends response.audio.done / response.done while the connection remains open, open a Microsoft Azure support case with the trace and correlation IDs because that indicates an Azure service-side streaming defect rather than a normal client implementation issue.

Use the below resource links for more reading and implementation steps:

I hope this is helpful. Please! Do not hesitate to let me know if you have any other questions, steps or clarifications.

Please do not close the thread by upvoting and accepting the answer if any part of it is helpful.