Video is now required for WebRTC realtime connections?

My demo apps started failing with the error below. Seems video is now required for realtime WebRTC connections. Is there any info about this anywhere?

Failed to parse SessionDescription. {"error":{"message":"Invalid SDP offer. Offer should have 1 video media sections, but had 0","type":"invalid_request_error","param":null,"code":"invalid_offer"}}
5 Likes

same issue - i think this is just a bad merge from openai side

3 Likes

Same issue here.

It looks like the changes haven’t been applied to all instances though, as some requests do succeed.

Forget what I said about some requests succeeding, looks like the deploy of this bug is complete. :wink:

Hoping this gets resolved reasonably quickly.

1 Like

Same here.
Adding a dummy video transceiver (example) const YOUR_RTC = new RTCPeerConnection();

.......
YOUR_RTC.addTransceiver('video', { direction: 'recvonly' }); resolves the realtime api connection  problem, but then I'm not receiving audio deltas from the model.  I'm only getting the text data correctly.

Same here. Came back after a few hours AKF and get this error too, no code change on our side.

it’s working again, back to building!

Working again for me without video as well.

I did do a test and added a video track and was hoping to get an incoming video stream but alas no packets. Hopefully the avatars are coming soon.

Actually there is at least one subtle change compared to the version from a couple hours earlier. When there is an audio response, it starts with the output_audio_buffer.audio_started event and ends with output_audio_buffer.audio_stopped. After the issue with the video track, it is now output_audio_buffer.started and output_audio_buffer.stopped

Yes, it was, sorry — the fix went out ~3 hours ago and Realtime should be back to normal.

1 Like