I have these two calls where I tested different combinations of chunkPlan, and in both cases I get the same result. The llm model returns a single message "output": " Confirma que a matrícula do seu veículo é um, dois, três, quatro, á, bê?" but VAPI gets sent in two independent requests to the voice model: "input": "Obrigado pela correção, Tiago. Confirma que a matrícula do seu veículo é um, dois, três, quatro, á," and "input": " bê?" This causes a long pause that negatively impacts the conversational experience. How can I prevent vapi from splitting it into two messages? I tried disabling chunkPlan in this call 50132580-dfc8-4060-9db5-bed0bfb6eefe "voice": { "voiceId": "pt-PT-FernandaNeural", "provider": "azure", "chunkPlan": { "enabled": false }, "cachingEnabled": false } In another call I tried with punctuationBoundaries and it didn’t work either. The result is always the same 13dd7aae-237d-415e-a692-acff62df881c "voice": { "voiceId": "pt-PT-FernandaNeural", "provider": "azure", "chunkPlan": { "punctuationBoundaries": [ ".", "!", "?", ";" ] }, "cachingEnabled": false }

I understand the issue, that split into two requests is exactly what causes the long pause. Disabling chunkPlan or tweaking punctuation boundaries alone won’t stop it. I can help you adjust the voice streaming so the whole sentence is delivered as one chunk without cutting mid-utterance. Have you tried enabling singleUtterance mode or tuning the buffer/latency settings on the VAPI side? If you’d like, we can go through your setup together and get it running smoothly. @david.jpi

@Vapi Support Bot @Vapi

To help prevent Azure from splitting your message, try explicitly enabling chunkPlan with a high minCharacters value to force the entire message to be sent as one chunk. For example:

json
"voice": {
  "voiceId": "pt-PT-FernandaNeural",
  "provider": "azure",
  "chunkPlan": {
    "enabled": true,
    "minCharacters": 500
  },
  "cachingEnabled": false
}

This should reduce mid-utterance splits by ensuring the chunk is large enough to include your full sentence. Test this configuration and see if it resolves the pause issue. (Reference: VAPI documentation on chunkPlan configuration) Source: - [VAPI Voice Pipeline Configuration](https://docs.vapi.ai/customization/voice-pipeline-configuration)

minCharacters have a maximum threshold of 80.

Hi David Jurado, Yyou can control voice transmission timing using several methods[1](https://docs.vapi.ai/assistants/flush-syntax): 1. Use flush syntax to control exact transmission points:

"Obrigado pela correção, Tiago. <flush /> Confirma que a matrícula do seu veículo é um, dois, três, quatro, á, bê?"

2. Configure voice formatting plan:

{  "voice": {    "chunkPlan": {      "enabled": true,      "strategy": "sentence",  // Try this instead of punctuationBoundaries      "maxChunkLength": 200   // Adjust this to prevent unwanted splits    }  }}

3. Best practices for chunking: - Place breaks at natural sentence boundaries - Avoid mid-sentence splits - Keep chunks reasonably sized for your voice provider If you're still experiencing unwanted splits, try: - Increasing

maxChunkLength

- Using the

sentence

strategy instead of

punctuationBoundaries

- Adding explicit

<flush />

controls where needed