Unable to Interrupt Model in Realtime API Using ServerVAD

[marianafdz465]

Hi!
I’m currently implementing the Realtime API using the OpenAIRealTimeSample scene provided in the SDK. Everything is working as expected except for the ability to interrupt the model while it is speaking.
I’ve added the turnDetectionSettings when creating the SessionConfiguration, as shown below:

new SessionConfiguration( model: Model.GPT4oRealtime, instructions: systemPrompt, turnDetectionSettings: new ServerVAD( createResponse: true, interruptResponse: true ), voice: Voice.Verse, tools: tools )

Despite this configuration, the model continues speaking and does not stop when I start talking. Is there something else I need to implement on the client side to enable interruption?
Thanks for your help

[StephenHodgson]

Hi @marianafdz465, I've not personally tried out these features myself. But did you send another user message to the model? I think it should also interrupt if the user sends any speech and it is detected as well.

Do you have some debug information or a repro steps I can take to test this out myself?

[marianafdz465]

Thanks for your response.

To help reproduce the issue:
I’m using the sample scene called "OpenAIRealTimeSample", located at:
Assets/Samples/OpenAI/8.6.6/Realtime

I imported this by installing the Realtime samples from the OpenAI 8.6.6 package available in your registry.

The only change I made was modifying the RealtimeBehaviour script to include a ServerVAD configuration in the SessionConfiguration. As a result, I am able to receive audio responses from the agent, but interruptions are not working , the model continues speaking even when I start talking or I asked to stop.

[StephenHodgson]

I can take a look and see