All posts
October 15, 2026
Realtime or cascaded: choosing a voice engine
Speech-to-speech models answer fastest. A cascaded pipeline lets you pick any model. Here is how to choose.
Purple Voice offers two ways to run an agent.
Realtime
OpenAI's Realtime models hear and speak directly. Replies start in a few hundred milliseconds and interruptions feel natural. Choose it when conversational feel matters most.
Cascaded
Speech is transcribed, the text goes to the model you choose — Claude, Gemini or OpenAI — and the reply is spoken back. Choose it when you want a specific model's reasoning or writing style.
You can switch an agent between engines at any time and compare results in Insights.
- product
- ai-agents