When creating a websocket connection using the wss://api.openai.com/v1/realtime endpoint, and trying to update the session data with the session.update event like it was working before the recent gpt-realtime update, it returns an invalid_request_error: Unknown parameter: ‘session.audio’ regarding the format which follows the documentation to the letter.
Refer to: https://platform.openai.com/docs/guides/realtime-websocket for websocket initiation with session.updateon open
And: https://platform.openai.com/docs/api-reference/realtime_client_events/session/update for the session.updateclient event format
The same error is returned when specifying the session.type field