Hi @Vapi @Vapi Incident @Vapi Support Bot . We're experiencing a Transcriber issue that is impacting our call experience. We're using Deepgram + Flux. When the call starts, it looks an issue is happening between Vapi Server and Deeppgram... I saw this error here in the Vapi Call logs. > Error: Unexpected server response: 400 at ClientRequest. (/app/node_modules/ws/lib/websocket.js:913:7) at ClientRequest.emit (node:events:532:35) at ClientRequest.emit (node:domain:488:12) at HTTPParser.parserOnIncomingClient (node:_http_client:704:27) at HTTPParser.parserOnHeadersComplete (node:_http_common:118:17) at TLSSocket.socketOnData (node:_http_client:546:22) at TLSSocket.emit (node:events:520:28) at TLSSocket.emit (node:domain:488:12) at addChunk (node:internal/streams/readable:559:12) at readableAddChunkPushByteMode (node:internal/streams/readable:510:3) at Readable.push (node:internal/streams/readable:390:5) at TLSWrap.onStreamRead (node:internal/stream_base_commons:191:23) at TLSWrap.callbackTrampoline (node:internal/async_hooks:130:17) We have a fallback in place, but this error is impacting the call experience since the assistant become unresponsive when the call starts. So, the callers are experiencing a terrible experience. Do you have any idea on what is happening?

That 400 error usually means the Deepgram stream isn’t initializing correctly, often a mismatch in model, language, or audio settings, so the assistant never fully connects at call start. I’d double check the Vapi Deepgram config (model, sample rate, encoding) and try hitting Deepgram directly to isolate it. This is likely why the assistant goes silent right when the call begins. Happy to help you fix this end to end are you overriding any transcriber settings on call start? @Leo

Thanks @Matt . This is intermittent. Some calls are working fine, while other don't... We're using the same model for all calls.

{
  "model": "flux-general-en",
  "language": "en",
  "provider": "deepgram"
}

Thank you for letting us know the issue is fixed