I am facing a problem, where there is some silence in between a sentence (The bot skips 3-4 words) sometimes, and there is significant lag sometime in the response. A couple of examples Call Ids : 019bdc66-b426-7aa9-890c-484e8ff717b2 - See the message "Sure. Type 2 diabetes does increase a lot of risks. Especially cardiovascular. But most of my patients are pretty well controlled on metformin. Where are you going with this?" and listen to the audio of same 019bc7e9-64f0-7003-b7bf-6bf30884dc75 - See the message "Yes, Matt. I remember. We talked last week about NIVEXIA's shot formulation, the subcutaneous version of Nivexia. Right? Had a lot of concerns about absorption and long term efficacy, especially in my late stage cancer patients who've been stable on ket a v for years. You said you'd bring more data and specifics. So what do you have for me?" and listen to the audio of the same

To reduce lag and prevent voice breaking, try adjusting the

waitSeconds

parameter in your

startSpeakingPlan

to 0.4 seconds, which is optimal for standard conversations. Here’s a sample configuration:

json
"startSpeakingPlan": {
  "waitSeconds": 0.4
}

Apply this setting and test if the response flow improves. For further tuning, refer to the [voice pipeline configuration guide](https://docs.vapi.ai/customization/voice-pipeline-configuration). Source: - [Voice pipeline configuration](https://docs.vapi.ai/customization/voice-pipeline-configuration)

@Vapi Support Bot its already set to 0.4.

Thank you for confirming that your

waitSeconds

parameter is already set to 0.4. I do not have additional troubleshooting steps or technical details available beyond what has already been discussed. If you have made this adjustment and are still experiencing issues, I understand how frustrating that can be. If there are any other details or changes you’ve noticed, feel free to share them!

@VAPI Tech can you help debug this issue?

hey, i faced similar issue. I've DM'd you

This usually happens when audio chunks arrive late or out of order, causing brief silence that sounds like skipped words even though the text is correct. The response lag is typically upstream latency from STT endpointing, model generation, or any tool/webhook calls tied to that turn. Streaming TTS is more prone to this, especially under network jitter, so gaps can appear mid-sentence. If interruptions or aggressive silence detection are enabled, the bot may cut and resume its own playback, creating audible holes. Slow webhooks or tools that don’t return a quick 2xx can also introduce pauses before audio resumes. Using larger responses or slower voice presets increases the chance of these artifacts. Buffering the audio better, simplifying the flow, or switching to non-streaming TTS usually reduces both the silence and lag.

Hey — I looked into your call 019bdc66-b426-7aa9-890c-484e8ff717b2. Root cause: brief interruptions in the [Daily.co](http://Daily.co) WebRTC audio stream are triggering silence injection, which creates the mid-sentence gaps you’re hearing. TTS (ElevenLabs) generated full audio correctly; the gaps are happening at the transport layer due to jitter/packet loss. What to do: • Check network stability on both ends (WebRTC is sensitive to jitter). • For web calls, make sure the browser isn’t CPU-throttled and avoid unstable Wi-Fi. • Optional mitigation: start playback sooner by lowering the chunk size:

"voice": { "chunkPlan": { "minCharacters": 15 } }

This is a network/transport issue (not TTS config), so gaps may persist if conditions are unstable.

no the problem is not in the transport layer of daily server to client SDK, it is from Vapi server to Daily SFU I have describe in depth here https://discord.com/channels/1211482211119796234/1468504819110252584 reason why i am confident is because vapi dashboard audio is clear but daily recording is same as call experience (jitters)

we are currently investigating this issue and will get back to you with an update shortly