Turn-taking and barge-in

Learn in five minutes

• LiveKit's turns page says turn detection decides when the user has finished speaking and when they start speaking while the agent is talking.
• That page says most techniques use voice activity detection, then endpointing to decide the end of a phrase.
• The same page describes adaptive interruption handling as a way to separate a real interruption from backchanneling.
• The xAI Speech to Speech page says turn_detection.type set to server_vad detects when the user has finished speaking.
• That page also says to enable server_vad for automatic barge-in, and that item.interruptible set to false drops caller audio until playback completes.

Barge-in in this lab means the person speaks while the reply is still playing. Copy the field name from the page you opened. Do not invent a silence timer.

This section HTML

Visual

Diagram SVGDiagram PNGDiagram HTML

Flow SVGFlow PNGFlow HTML

Two ways to yield

1. Write the LiveKit sentence: the agent can yield when the user starts speaking mid-response.
2. Write the xAI sample field: turn_detection.type is server_vad.
3. Write the xAI field that drops caller audio: item.interruptible false.
4. Label each line with the page it came from.

This section HTML

Quiz

This section HTML

Ecosystem

Barge-in

The person speaks while the assistant is still talking. Vendors name the detector differently.

Sisters: Speech-to-text, Language model, Text-to-speech, WebRTC, SIP.

A cascaded pipeline and a speech-to-speech model are two shapes the LiveKit docs name. The xAI Speech to Speech API is a hosted speech-to-speech session.

Pin the model id, the voice, and the date you read the docs.

Config

• Write the goal in one sentence.
• Name the input and the allowed output.
• Save the trace next to the result.

Troubleshooting

• The answer looks fluent and is wrong. Check whether a source passage actually contains the claim.
• The loop never stops. Set a step cap and stop when the tool result repeats.

This section HTML