Using realtime with turn detection disabled (null) and sending response.create after committing input audio buffer does not generate audio or text anymore, and just immediately returns response.done with no errors.
This seems to have bugged out since 23/01/2025, there was no code change and production builds broke over night.
The transcripts and user’s audio detection is all correct and nothing seems to be broken on that end.
I thought maybe this was related to the introduction of create_response under turn detection object, but responses are not working with create_response: false either.
Even though this may appear offtopic, @robson.costa, what happens if you try to render any pic in chatgpt, for example:
render a dog pic, is this also stuck?
Because for me the same happens as for you, but also for rendering pics.
I am asking because: Potentially we’ll face another outage of OpenAI functions, like ChatGPT and others already again.