Hello Vapi Support, We have a recurrent issue on PSTN inbound calls where the caller is clearly speaking (audible in the call recording waveform), but Vapi does not create a new user turn and the transcriber shows no transcripts/events during that segment. The assistant stays silent for a long time (e.g., 50+ seconds). Call ID: 019c143f-4c46-7ee5-aac5-f12d774ff3d3 Timestamp: around minute ~3:00 the caller repeatedly says “hola/hello”, audible in the recording waveform, but there are no transcriber/endpointing user-turn events. Additional details: In the dashboard, downloading “PCAP” only provides a sip.pcap (SIP signaling only). We need a full PCAP including RTP to verify inbound audio and packet stats (jitter/loss) during the failure window. Logs show voice (ElevenLabs) websocket activity, but no transcriber activity during the caller speech segment. Could you please: Confirm whether inbound RTP audio was received by Vapi during the failure window (and provide jitter/loss/packet stats if available). Explain why the VAD/endpointing did not detect speech / create a user turn even though the caller is audible. Provide (or enable) a downloadable PCAP including RTP for this call (not SIP-only). Thanks.

Hi there, Thank you for your message. Our team is currently out of the office. We operate Monday through Friday, from 9:00 AM to 8:00 PM Pacific Standard Time (PST). We’ll get back to you as soon as possible during our normal business hours. If your message is urgent, please mark it accordingly or include “URGENT” in the subject line, and we’ll do our best to respond promptly. Warm regards, Vapi Customer Support Team

Suggestion: Since Vapi's orchestration layer uses advanced endpointing and background voice filtering, it's possible that the caller's speech was filtered out or not detected as a user turn, even though it was audible in the recording. To further diagnose, review your assistant's speech configuration—especially the endpointing and background noise/voice filtering settings—to ensure they are not overly aggressive. You can adjust these parameters via the API to better capture user speech in noisy or variable environments. For more details on how these models work and configuration options, see the [Vapi orchestration and speech configuration documentation](https://docs.vapi.ai/how-vapi-works). Source: - [How Vapi works](https://docs.vapi.ai/how-vapi-works)

Please Need support

Hi Estefan, Thanks for your patience while we reviewed this. We’ve analyzed the logs for Call ID: 019c143f-4c46-7ee5-aac5-f12d774ff3d3 and can confirm the following: What we found Inbound RTP audio was successfully received during the failure window (a total of 11,077 RTP packets). During the ~3:00 mark, the VAD (Voice Activity Detection) did detect changes in audio energy, which were logged internally as “Speech Possibly Starting”. However, the audio never crossed the confidence threshold required to confirm speech. Because speech was never confirmed: No audio was forwarded to the transcriber No user turn was created The assistant remained silent Root cause This behavior typically occurs when the caller’s audio amplitude is below the VAD detection threshold, which can happen due to: - Low audio levels on the SIP trunk - The caller speaking very quietly - Background noise affecting signal clarity Even though speech is audible in the recording, it may not be strong or clean enough to pass VAD confidence checks in real time. Recommended next steps To further diagnose and mitigate this: - Review the stereo recording around ~3:00 to assess the caller’s audio amplitude during the failure window. - Check and adjust audio gain settings on your BYO SIP trunk to ensure adequate inbound levels. - Add idle prompts so users are proactively prompted when speech isn’t detected. For example:

{
  "idleMessages": [
    "Are you still there?",
    "I didn't catch that. Could you repeat?"
  ],
  "idleMessageMaxSpokenCount": 3,
  "idleTimeoutSeconds": 10
}

This helps recover conversations when speech is present but not confidently detected.