Call ID: 019d259e-3913-7000-927e-f4dc6a33fae5 Time: 2026-03-25 15:30 UTC The call was configured with firstMessageMode: "assistant-speaks-first" and a firstMessage. Three ElevenLabs WebSocket connections opened and sent BOS successfully, but zero text messages were ever sent to any of them — no first message, no hook timeout prompts. The caller heard complete silence for the entire call. The Deepgram transcriber also took 22 seconds to open (vs the normal <1s). Speech-update webhooks were sent as if the assistant was speaking, but no audio was actually produced. No errors were logged in the call log. All webhooks returned 200. The assistant config and voice settings are correct — other calls with the same config work fine. Can you help me understand what happen or if that can happen again?

This was a transient Vapi infrastructure (TTS pipeline) failure. To help prevent silent calls in the future, configure a fallback TTS provider in your assistant setup:

json
{
  "voice": {
    "provider": "elevenlabs",
    "voiceId": "your-voice-id",
    "fallbackPlan": {
      "voices": [
        { "provider": "azure", "voiceId": "en-US-JennyNeural" }
      ]
    }
  }
}

If the issue persists, contact Vapi support with your call ID for deeper investigation. [More info](https://docs.vapi.ai/calls/troubleshoot-call-errors) Sources: - [Troubleshoot call errors](https://docs.vapi.ai/calls/troubleshoot-call-errors) - [Call ended reason](https://docs.vapi.ai/calls/call-ended-reason) - [Discord community discussion](https://discord.com/channels/1211482211119796234/1370485222214336564)

This looks like the call flow got stuck TTS connected, but the first message never actually went out, so nothing triggered audio even though it showed “speaking.” The 22s Deepgram delay also feels like a cold start or brief stall that blocked things at the start. I’d double check if the first message event actually fired or got dropped before reaching ElevenLabs. Has this only happened on this one call, or have you noticed a few silent calls like this recently? @Filippo

Only this one instance.

Got it since it only happened once, it was likely just a temporary stall where the first message didn’t go out, so audio never started. The Deepgram delay also suggests things were slow to spin up at the start. I’d check the event timeline (first message → TTS → transcriber) to see where it got stuck. It’s also worth adding a simple fallback to resend the first message if no audio plays after a couple seconds. @Filippo

Ok, will check if happens again and then add fallback