Respond only with text and without audio ，when the conversation.item.create event sends assistant text messages to populate the conversation history.

[hsycc]

like this

{
    "event_id": "event_345",
    "type": "conversation.item.create",
    "previous_item_id": null,
    "item": {
        "id": "msg_001",
        "type": "message",
        "role": "assistant",
        "content": [
            {
                "type": "text",
                "text": "Hello"
            }
        ]
    }
}

[hsycc]

see the document, has the current limitation that it cannot populate assistant audio messages.

Is the lack of audio response at present a bug?

[LaKing]

I would assume that modalities: ["audio", "text"] for the client session would adjust session-wide, but it seems not to, in fact removing audio makes the api or the model just hang.

It would be nice to find a way to set if we need audio responses or not. ...

[guvenkaranfil]

Hello @hsycc ,
We are having the same isssue. Have you find a solution?

[hsycc]

Hello @hsycc , We are having the same isssue. Have you find a solution?

@guvenkaranfil Sorry, we still haven't resolved this issue. Regarding the problem with the GPT model itself, I haven't paid attention to it for the past ten months. If you want to send the chat context to the LLM for processing, you can try configuring this in the system prompt