After about 20-30 seconds of interaction with the Realtime Audio API, the connection drops. My audio is still physically transmitted from my side, but I stop receiving responses from the server at that point. It’s like the connection breaks mid-sentence from their end, cutting off the reply. This issue is consistently reproducible. Please help troubleshoot this.
You are probably hitting the max tokens. When this happens I will see the full text in transcription but the model stops sending. .. no other warning, it just stops!
be sure MAX_OUTPUT_TOKENS is set high enough… I need a few thousand for a friendly conversation about news, family, weather, etc. My application is a companion for senior citizens.
{
"type": "response.create",
"response": {
"conversation":"auto",
"instructions": greet_user,
"max_output_tokens": MAX_OUTPUT_TOKENS,
"output_modalities": ["audio"],
"audio": {
"output": {
"format": {
"type": "audio/pcm",
"rate": NATIVE_OAI_SAMPLE_RATE_HZ
},
"voice": master_state.conman.get_config("VOICE"),
}
}
}
Hey, thanks a lot for the suggestion - I actually checked that too, but it turned out not to be a token issue.
My logs showed that audio was still streaming normally while the DataChannel suddenly closed in the middle of the session.
Once I fixed the DataChannel handling and reconnection logic, everything started working perfectly - no more 20–30s dropouts.
Appreciate the help though, it really pointed me in the right direction