The voice loop

Learn in five minutes

• A voice turn is: hear, understand, decide, speak.
• Hearing and speaking are models or browsers of their own, not the language model.
• The Web Speech API is documented by MDN. The W3C draft should be confirmed. It is not enough to call it a Recommendation.
• Barge-in means the person speaks over the reply.
• The loop still has a tool cap.

If the browser both hears and speaks, the language model only sees text. Draw that split so you do not treat the microphone as the LLM.

This section HTML

Visual

Diagram SVGDiagram PNGDiagram HTML

Flow SVGFlow PNGFlow HTML

Four boxes

1. Mic to text.
2. Text to the same agent loop as chat.
3. Tool or refusal.
4. Text to speech.
5. Label the speech boxes as separate from the language model.

This section HTML

Quiz

This section HTML

Ecosystem

Barge-in

The person starts speaking while the assistant is still talking.

Sisters: Web Speech API, Turn-taking, Queues, CRM notes.

The retrieval index and the tool cap are shared. The channel changes the wait and the consent copy.

A later host could add a channel without rewriting the notes server.

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