Artificial Intelligence · 1 min read
Speech-to-Speech Model for Voice AI Agents: Setup and Test Guide
Voice AI offers a full-duplex speech-to-speech model. The agent can listen while it speaks, so callers can interrupt naturally and the agent adjusts right away instead of waiting for a turn to end.
The model splits the work in two:
- The speech part handles the live conversation.
- The reasoning part handles decisions and actions, such as booking or transferring.
Before you start
- You need permission to view and manage Voice AI agents.
- Set up the actions your agent needs, such as booking or call transfer.
Set it up
- Go to AI Agents > Voice AI and open an agent.
- In the agent's model settings, choose the speech-to-speech model.
- Write the speech instructions: greeting, tone, and how to talk to callers.
- Write the reasoning instructions: when to take each action and what information to collect.
- Save the agent.
Test it
- Start a test call.
- Interrupt the agent mid-sentence and check that it stops and responds.
- Ask for each action, such as booking an appointment, and check that it happens.
- Review the call log and transcript.
Keep speech instructions short and conversational. Put business rules in the reasoning instructions.