Hello, We’re experiencing a similar issue again and would like to reopen the support case, as the problem seems to have resurfaced. Previously, we reported that the model returned a response, but it was never played, and the call ended with the reason "Silence Time Out." Now, we’re facing the same behavior. Here are two recent call IDs where this occurred: efe77b7b-9a58-435a-8245-55e81f20f7a4 247cd920-068d-4e82-8bc4-0e2fa4f8fee9 Could you please provide an update on this issue or help us investigate further? Thank you.

Thanks for sharing the details and call IDs! I checked our resources, but I couldn't find enough info to fully answer your question about the 'Silence Time Out' issue where the model returns a response but it's not played. I've escalated your case to a human support specialist who can dig deeper and provide an update or next steps. For reference, 'Silence Time Out' means the call ended due to prolonged silence or inactivity—more details are available [here](https://docs.vapi.ai/calls/call-ended-reason). We'll get back to you as soon as we have more info!

Try increase silenceTimeoutSecond..

Hello, please, I need you to look into this issue in depth, because in our previous case @Shubham Bajaj told us that it was an error on your side and that you were working on it. https://cdn.discordapp.com/attachments/1383395050305617940/1383418714229313597/image.png?ex=684eb894&is=684d6714&hm=8d721d43f248f5362d15441dc2e6fecc674e7293b36634cf6adc94b04a8de2e9&

Hey David, can you please use gpt-4o instead of realtime cluster? It will solve the issue.

Hello Sahil, We are currently using a custom LLM solution. The problem we're facing is that, although we send back the response in the appropriate format and within the expected time, the response is never delivered to the customer in the form of voice. As a result, the customer hears nothing and waits for the assistant's reply. Eventually, the call ends due to a silence timeout.

I am having the same issue. Were you able to resolve this?

@Shubham Bajaj @nikhil Here is another call with this error (callId: c216360b-65d5-4f38-bc2f-e18f4be471dc). We urgently need a response — this issue is seriously affecting us and causing major problems with our customers.

Can you use gpt-4o instead of realtime?

Hello Sahil, we are not using realtime. We use a custom-llm solution that it use gpt-4o

Can you please confirm if your custom-llm server url is streaming the responses? Vapi by default expects server-sent events (SSE)

Hello again @Shubham Bajaj @Kings_big💫 @User , We are intermittently facing the same issue. It occurs in some calls but not in others. What’s common across all cases is that our custom LLM responds correctly, yet the audio is never heard by the user. I've seen that this is a problem other clients have also experienced, and it hasn’t received a proper response. We kindly ask you to investigate this thoroughly, as it’s not a configuration issue — it appears to be a random error on VAPI’s side. This is causing a poor user experience and abrupt call disconnections. Our clients are raising concerns, asking why this happens during calls, and we are currently unable to give them a clear explanation. Moreover, this issue has been happening for several months now, and we’ve reported it multiple times from our account. We would really appreciate it if you could analyze it in depth and provide us with a detailed explanation of what might be going on. CALL ID: ab33e46e-3cbf-429d-94ab-e11d6d967579, e9dcfbea-754a-406c-8915-dbe9b66d8080 Thank you very much.

Hello @User @Sahil i'm facing this problem too, it's happening pretty often and the conversational experience turns instantly bad because the assistant shut up and the customer doen't know how to handle this and hangs the call, please can you have a look at this problem?? thank you!

After reviewing the logs, what happened was the customLLM responded, but the websocket to elevenLabs voice was not ready and dropped the say request, resulting in no speech.

🟡 10:58:55:791 ElevenLabs (Websocket #4) State "connecting" But Received End Token. Dropping 143... "Perfecto. Le atenderemos en nuestras instalaciones de Sant Cugat, ubicada en Carretera de Rubí, setenta y seis. Dígame, ¿en qué puedo ayudarle?"

To prevent this issue in the future, you could configure some fallbackVoices in case the connection to ElevenLabs fails.