https://vapi.ai/ logo
Join Discord
Powered by
Web Call assistant cuts off when user starts speak...
# support
v
Hi — need help debugging an issue with my web call demo. Issue: Web Call Vapi assistant cuts off the moment the user starts speaking. The call connects, the assistant delivers its greeting, but as soon as the user begins responding, the call drops with no further communication. Reproduces consistently every attempt. Call ID (just reproduced): 019dcec2-0cca-7001-a6f2-87cb208fd4de Assistant ID: 40fede95-513c-4ddf-8e80-7360c7391741 Domain: demo.talktoporter.com Use case: Sales demo for prospective customers (AI receptionist for service businesses) Email ticket from Apr 24 got an auto-acknowledgment but no human follow-up after 3+ days, hoping Discord moves faster. Happy to provide more reproduction details, browser console logs, or test from a different network/browser. What additional info would help? Thanks.
p
Hey this usually isn’t the call itself but how the audio handoff is configured. When the assistant finishes speaking and the user starts, something in the stream (often barge-in or mic/WebRTC handling) is breaking and ending the session. I can take a look at your setup, reproduce it, and pinpoint exactly where it’s failing so your demo runs smoothly. If you can share your config or console logs, I’ll sort it out quickly @vonGimbel
v
Reproduced fresh with browser console open. New Call ID: 019dcf02-ef42-7002-8324-0879e0ed7752 (assistant: 40fede95-513c-4ddf-8e80-7360c7391741). Attached: - Voice tab config - Transcriber tab config - Advanced tab config (including client/server messages) - Browser console screenshot from the fresh ejected call Same symptom: Porter delivers greeting, the moment I start speaking the call cuts off and there's no further communication. Let me know what else helps. https://cdn.discordapp.com/attachments/1498291188300517486/1498307902761599066/Screenshot_2026-04-27_at_6.56.41_AM.heic?ex=6a067039&is=6a051eb9&hm=1b0c4e3aca4050dbf81161bd05c832eb6cbc0a667967bedc4d6b83f66afdd028& https://cdn.discordapp.com/attachments/1498291188300517486/1498307903344349225/Screenshot_2026-04-27_at_6.48.57_AM.heic?ex=6a067039&is=6a051eb9&hm=b32ba6d9a85d5ae0ce74e1e010142fb6cdcebfe59f9781f4f5430c50d4a0a3cd& https://cdn.discordapp.com/attachments/1498291188300517486/1498307904217022555/Screenshot_2026-04-27_at_6.47.49_AM.heic?ex=6a067039&is=6a051eb9&hm=b066bf2e5d1d7cc8c754642ad589ab7b9e5b5e2ce267a351f1a1898ea89152b5& https://cdn.discordapp.com/attachments/1498291188300517486/1498307905202684115/Screenshot_2026-04-27_at_6.47.25_AM.heic?ex=6a067039&is=6a051eb9&hm=d8cffe63a4aa3910286aa1fefd5883ea85e3f0ddc3b2cc79ca3ed567fe85c29a& https://cdn.discordapp.com/attachments/1498291188300517486/1498307906074972170/Screenshot_2026-04-27_at_6.47.13_AM.heic?ex=6a06703a&is=6a051eba&hm=60d5f5f45ddb1121443d93bdfbc666f8e798d31a4d9beddddf2f23813634ed63& https://cdn.discordapp.com/attachments/1498291188300517486/1498307907207561257/Screenshot_2026-04-27_at_6.45.16_AM.heic?ex=6a06703a&is=6a051eba&hm=85a082abb543625d1eff85875a3f0e7d854dd454ae83f881df9ef733d8f30eea& https://cdn.discordapp.com/attachments/1498291188300517486/1498307908092432588/Screenshot_2026-04-27_at_6.45.07_AM.heic?ex=6a06703a&is=6a051eba&hm=a28880ea53c65e818505287796b8920f1f38e6ffc838998e511ec241f73d243a& https://cdn.discordapp.com/attachments/1498291188300517486/1498307908880830586/Screenshot_2026-04-27_at_6.44.49_AM.heic?ex=6a06703a&is=6a051eba&hm=987e10d23364dfc1b00437eda01706647069e21988fc1058ddebfd34bd632001& https://cdn.discordapp.com/attachments/1498291188300517486/1498307909770281152/Screenshot_2026-04-27_at_6.44.15_AM.heic?ex=6a06703a&is=6a051eba&hm=2f68ed2c21f5d1bcad85bc5af9e33e330a3cce6643e00cf82e974264b93629d4& https://cdn.discordapp.com/attachments/1498291188300517486/1498307910785175675/Screenshot_2026-04-27_at_6.43.48_AM.heic?ex=6a06703b&is=6a051ebb&hm=79fa497358de39e1fb3ce4bb5a0bf42f55c588a856d018783a1a014aa2cb9087&
p
Ohhh the console makes it clear. Daily.co (Vapi's WebRTC layer) is trying to load an audio input processor your browser doesn't support, which ejects the call the moment you start speaking. Switch to Chrome on desktop if you aren't already, and disable noise cancellation in your Vapi Voice settings if it's on. Also drop Input Min Characters from 30 to around 10 won't fix the disconnect but the assistant will feel more responsive once it's sorted. Let me know what browser/OS you're on and I'll dig deeper if needed. @vonGimbel
v
Thank you — that makes sense. I'm on [Chrome/Safari/Firefox] on [Mac/Windows]. Going to disable noise cancellation, drop Input Min Characters to 10, and confirm I'm on Chrome desktop. Will report back.
Made the changes — disabled noise cancellation in Voice settings, dropped Input Min Characters to 10, saved the assistant config. Reproduced fresh on Chrome desktop on [Mac/Windows]. Same symptom: greeting plays, the moment I start speaking the call cuts off. What's next?
019dd0ff-2234-7002-a217-a71d2bf772b6
s
Hi, I’ve investigated call 019dcec2-0cca-7001-a6f2-87cb208fd4de and found the root cause. What happened The call isn't actually "cutting off" - it is being terminated because your custom LLM server is crashing.
The Greeting: The assistant’s first message works because it is pre-configured and doesn't require a call to your server. The Drop: The moment you speak, Vapi sends your transcript to your configured URL: https://sheshq.talktoporter.com/vapi/demo/chat/completions. The Error: Your server immediately responded with a 500 Internal Server Error. Because Vapi cannot get a response from your "brain" (the LLM), it has no choice but to end the call.
How to fix it The issue is on your backend. You can verify this by running a simple test from your terminal:
Copy code
bash
curl -X POST https://sheshq.talktoporter.com/vapi/demo/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "porter-demo",
    "messages": [{"role": "user", "content": "hello"}],
    "stream": true
  }'
If this returns an error, you'll need to check your server logs to see why the code is crashing. Common culprits are missing environment variables, expired API keys, or unhandled exceptions in the request body.
Once your server starts returning valid 200 responses, the calls will work immediately without any changes needed on the Vapi side.
v
Update — investigated further. Server side is healthy: I can confirm the /vapi/demo/chat/completions endpoint returns 200 with a streaming response when called directly via curl from the VPS. Server logs show ZERO POST requests to that endpoint during the failed demo calls — meaning Vapi never even reached the LLM during those calls. The browser console keeps showing this on every ejected call: "Ignoring settings for browser- or platform-unsupported input processor(s): audio" Followed immediately by "Meeting has ended due to ejection: Meeting has ended" via Daily.co. This is on Chrome desktop on [Mac/Windows]. I previously turned off Background Sound in the assistant config. Are there other audio-input processor settings on the assistant or globally on my Vapi account that could be triggering this Daily.co audio rejection? My LLM looks fine — the eject is happening upstream of it in the WebRTC audio layer. Latest call ID: 019dd662-ebb9-7cc5-baf9-52c873b06f47
s
Hi, I took another look at your latest call (019dd662-ebb9-7cc5-baf9-52c873b06f47) and can confirm the root cause: The call is being terminated due to a 500 error from your custom LLM endpoint. What’s happening: The call connects and greeting plays (no LLM call needed) ✅ As soon as you speak, Vapi sends a request to: https://sheshq.talktoporter.com/vapi/demo/chat/completions That request is returning a 500 error, which causes the call to end Important clarification
The browser console message: "Ignoring settings for browser- or platform-unsupported input processor(s): audio" is a red herring — it’s just a Daily.co client-side warning and not the cause of the call drop. Why your server logs show “no requests”
This usually means the failure is happening at the nginx/proxy layer before your app, so: Vapi → nginx → ❌ fails nginx never forwards to your app → no logs on your backend
How to fix Please check: nginx error logs (not just app logs) Upstream service is running and reachable Correct port / proxy_pass config No missing env vars or crashing middleware Also worth noting: Your webhook endpoint is also returning 500s, which points to a broader backend issue Quick test Run this from outside your server:
Copy code
bash 
curl -X POST https://sheshq.talktoporter.com/vapi/demo/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "porter-demo",
    "messages": [{"role": "user", "content": "hello"}],
    "stream": true
  }'
If nginx returns 500 here, that confirms the issue.
Once your endpoint consistently returns 200, the calls will stop dropping immediately - no changes needed on the Vapi side. Happy to take another look once you’ve checked nginx logs.
PreviousNext
Powered by