Here is the setup for context: The call tool has a start message configured, and it is set to “Wait for message to be spoken before triggering tool call.” However, while the start message is being spoken, the audio suddenly cuts into a loud static sound before the tool call is triggered. Call ID: 019e2759-a354-7000-ba40-f5431e05565e

Hi Devin, To help track down this issue, could you share: - When exactly this happened (the timestamp) - What response you expected to get - What response you actually got instead This would really help us figure out what went wrong!

When exactly this happened: This happened atMay 14, 11:37:26.606 What response I expected to get: The response I hoped to get was for the tool to complete its start message which is "I would love to help you in achieving your goals. Our doctors would be happy to help. I can start the appointment request for you now by gathering a few quick details, and then our Patient Care Coordinator will reach out to confirm availability and help get your consultation scheduled." What response I actually got instead: however it got cut off in the middle of that message and just blasted static noise to my caller.

What happened to your caller: The audio cutoff + static noise was caused by the assistant's LLM returning an empty response and immediately calling the

handoff_pricing

tool — before any speech was generated. When the handoff fired, it abruptly terminated the audio stream (ElevenLabs V3), which created the static noise on the telephony side. The full start message was never actually spoken. ---- Immediate workaround for the user: They can add a

startSpeakingPlan

message to the

handoff_pricing

tool so the LLM is forced to speak before handing off. In their assistant configuration:

json
{
  "type": "function",
  "function": {
    "name": "handoff_pricing",
    "messages": [
      {
        "type": "request-start",
        "content": "I would love to help you in achieving your goals. Our doctors would be happy to help. I can start the appointment request for you now by gathering a few quick details, and then our Patient Care Coordinator will reach out to confirm availability and help get your consultation scheduled."
      }
    ]
  }
}

By adding a

request-start

message to the tool definition, the assistant will speak those words first before executing the handoff — ensuring the caller hears the full start message with no cutoff. ---- Why this works: The

request-start

message type is spoken synchronously before the tool call executes. This means audio is fully generated and delivered before any queue clearing or stream termination happens. ---- Longer-term platform fix (on Vapi's side): Our team is tracking a systemic fix for graceful audio fade-out on

assistantTransfer

events — but the workaround above will resolve this specific user's issue immediately.

The static in question did not occur when handoff_pricing was called. It happened two minutes later in the conversation.