Realtime API Not Responding

Hello, since around two days ago applications built using the Realtime API web sockets are ceasing to respond to user audio input. On debugging, OpenAI are sending back the following:

0|realtime-websocket | [VOICE-DEBUG] :cross_mark: ERROR FROM OPENAI: {
0|realtime-websocket | “type”: “error”,
0|realtime-websocket | “event_id”: “event_BB5o3FOLP6W1oyL9AXXXXX”,
0|realtime-websocket | “error”: {
0|realtime-websocket | “type”: “invalid_request_error”,
0|realtime-websocket | “code”: “invalid_value”,
0|realtime-websocket | “message”: “Audio content of 4800ms is already shorter than 6606ms”,
0|realtime-websocket | “param”: null,
0|realtime-websocket | “event_id”: null
0|realtime-websocket | }
0|realtime-websocket | }

This has happened across two separate applications using websockets, there were no code changes on our end, this seemingly happened overnight. This reproducible on the Twilio and OpenAI Realtime repository at GitHub - openai/openai-realtime-twilio-demo

Has anyone else encountered this and if so has anyone achieved a resolution? Thanks.

2 Likes

I wonder what requires 4.8 seconds of audio…maybe VAD noise profiling?

Given the internal message, is it happening just when your audio is short, or when it is just long? Maybe padding the length could be a workaround until this issue is addressed.

Hey there, thanks very much for the response. I have managed to resolve the issue, this might sound odd but we changed the telephone numbers on the Twilio end and it resolved the issue. I assume it’s an issue at one of the providers Twilio use, very strange, I hope this helps someone in the future.

1 Like