Live demo
Talk to my voice agent.
It listens, answers out loud and stops the moment you interrupt. Every answer shows how fast it came.
Transcript
- What you and the agent say shows up here.
Response time per turn
| Turn | Total | Turn detector | Model |
|---|---|---|---|
| Each answer adds a row: from the end of your speech to the agent's first sound. | |||
How it works
The hard parts, measured.
Streaming
Your microphone streams to a Python server over a WebSocket in 20 ms frames, and the reply streams back the same way.
Barge-in
Talk over the agent and it goes quiet at once. It also forgets the part of its answer you didn't hear.
Response time
Measured on the server, from the last moment of your speech to the first byte of the agent's audio. 400 ms of it is the silence the turn detector waits for before it decides you are done.
Speech model: OpenAI gpt-realtime-2.1-mini. One call lasts up to 3 minutes. Use headphones so the agent doesn't hear itself.