I'm so curious how are that natural stutters, pauses are implemented.

Vapi makes the Elliot voice sound natural by using a custom "filler injection" model that adds realistic stutters and pauses (like "umm", "like", "so") in real-time. This is handled by Vapi's orchestration layer, which transforms formal LLM output into conversational speech without extra latency. Example conceptual prompt:

Input: "Let me check that for you."
Output: "Umm, let me check that for you... so, just a moment."

For more details, see the [Vapi documentation on conversational models](https://docs.vapi.ai/how-vapi-works). Source: - [How Vapi works](https://docs.vapi.ai/how-vapi-works)

Hi, We have the details that help resolve your query in the documentation below. It provides a clear overview of how Vapi works, including the orchestration layer, endpointing, interruptions, and the overall voice pipeline flow. https://docs.vapi.ai/how-vapi-works