---
title: Use your voice in a thread
type: how-to
section: ai-studio
summary: "The Voice button next to the message box holds two modes: Live Chat, a spoken back-and-forth, and Dictate, which types what you say into the box."
applies_to:
  clients: [web]
sources:
  - evolve-front-end/app/[lng]/studio/components/voice/**
  - evolve-front-end/app/[lng]/studio/components/Chat/ChatInput.tsx
  - evolve-front-end/app/i18n/locales/en/powerflow.json
verified:
  at: 2026-09-04
  env: code
  by: content-lane-ai-studio
ttl_days: 45
related: [ai-studio/transcribe-an-audio-file, ai-studio/start-a-conversation]
---

Select the arrow beside **Voice** next to the message box and choose **Live Chat** to speak with the model, or **Dictate** to type what you say into the message box.

Your browser asks for the microphone the first time. Both modes need it.

## Speak with the model

1. Select the arrow beside **Voice**, then **Live Chat**.
2. Wait for **Voice is live**. Then speak.
3. Speaking while the model is talking interrupts it.
4. To finish, select the arrow and choose **Stop Live Chat**.

The exchange is saved as a thread titled **Voice conversation**.

If the live connection cannot be made, AI Studio falls back to **Transcript-only mode**: record a segment, then confirm it to put the text into the message box without spoken replies.

## Dictate into the message box

1. Select the arrow beside **Voice**, then **Dictate**.
2. Speak. Select **Stop Dictation** when you are done.
3. The text lands in the message box. Read it, correct it, then send.

## While voice is active

The **Tools** button next to the message box is unavailable while a voice session is starting or stopping. Wait for it to settle.

## If it did not work

- **Microphone blocked**: allow the microphone for this site in your browser's settings, then reload the page.
- **Voice unavailable in this browser**: use a current desktop browser. Voice needs audio interfaces some browsers do not provide.
- **Realtime voice is unavailable. Transcript-only mode is active.**: record segments and confirm each one, or type instead.
- **No speech was detected in the recorded segment.**: check the right microphone is selected in your browser and speak closer to it.