Using nodejs connecting from server side.
I’m running into this issue that some time around 5-10 minutes in - response.text.delta
server event starts to respond slowly and my websocket connection disconnects.
I then get error RealtimeAPI is not connected
from here: openai-realtime-api-beta/lib/api.js at main · openai/openai-realtime-api-beta · GitHub with eventName="input_audio_buffer.append"
.
Modifying the close
listener here: openai-realtime-api-beta/lib/api.js at main · openai/openai-realtime-api-beta · GitHub to ws.on('close', (code, reason)
I get code=1011
with reason="keepalive ping timeout"
Looking at the rate_limits.updated
event. I’m hitting neither the token limit cap, or the 15-minute session timer cap.
Is there another mechanism from OpenAI side that throttles requests and would cause this disconnect?
2 Likes
I figured it out—you need to monkey patch the api.js in the library with a keep-alive ping-pong mechanism. The base library currently doesn’t support long-running monologues well.
3 Likes
jill7
December 16, 2024, 7:59pm
3
Thanks for sharing - I am running into the same issue! I notice the connection closes after just 1-2 minutes.
Mind sharing your patch? I added the ‘ping’ but got an error as the event type is unsupported, so figure it’s missing the ‘pong’
Also, what are you noticing in practice is a duration for a “long running monologue” that’s unsupported?
jill7
December 16, 2024, 8:00pm
4
@hemlocket tagging in case it doesn’t notify
sps
December 18, 2024, 2:49am
5
The maximum session length has been increased from 15 minutes to 30 minutes on the Realtime API beta.
Here’s the latest announcement:
A bunch of big updates for the Realtime API today. We’re announcing support for WebRTC, meaning you can add speech-to-speech experiences with just a handful of lines of code . (A bunch of you asked after the Rudolph toy demo in the livestream for the embedded SDK — Sean has published it to Github here ). We’ve also released two new snapshots:
gpt-4o-realtime-preview-2024-12-17 , which has improved voice quality and more reliable input, and with 60%+ cheaper audio.
gpt-4o-mini-realtime-preview-20…
1 Like
@hemlocket - I’d love to know more about your patch, I’m running into the same issue on my end.