# How do you have a voice conversation with ChatGPT?

Date: 2023-10-03 · Author: Robert Scales · https://guidetoartificialintelligence.com/en/learn/have-voice-conversation-chatgpt/

**Answer:** If you subscribe to ChatGPT Plus or Enterprise, turn on voice conversations in the settings of the iPhone or Android app, tap the headphones icon and start talking. ChatGPT listens, answers out loud in one of five voices, and adds the transcript to your chat when you finish.

Voice conversations started reaching ChatGPT subscribers on September 25, in a rollout OpenAI is spreading over two weeks. If you have the feature, you can now talk to ChatGPT the way you would talk to a person on the phone: you speak, it answers aloud, you interrupt or follow up, and the exchange carries on without touching the keyboard.

Behind the scenes, your speech is turned into text by Whisper, OpenAI's speech recognition model, ChatGPT writes a reply as usual, and a new text-to-speech model reads it to you. What matters for a beginner is the practical side: how to switch it on, how to steer a spoken conversation, and what it is good for.

## Who can use it, and how to turn it on

OpenAI's help page for the feature sets out the basics:

| Question | Answer as of this week |
|---|---|
| Who gets it? | ChatGPT Plus and Enterprise users; it is a beta |
| Where? | The ChatGPT apps for iPhone and Android only, not the website |
| Which models? | GPT-3.5 or GPT-4 (Plus users have a GPT-4 limit of 50 messages every three hours) |
| Voices | Five to choose from |
| Required setting | Chat History & Training must be on, so the transcript can be saved |

To set it up:

1. Update the ChatGPT app, then open **Settings → New Features** and switch on **Voice conversations**.
2. Back on the home screen, tap the **headphones** icon in the top-right corner and pick a voice.
3. Start talking. ChatGPT listens, replies out loud, then listens again.
4. Tap to **interrupt** when it is talking, use **pause** if you need a moment, and tap **X** to leave. The full transcript then appears in the ordinary text chat.

On privacy, OpenAI says that during the beta your audio is sent to its Whisper service for transcription and not kept afterward. The written transcript is stored in your history like any other chat and, depending on your data settings, may be used to improve its models.

## A worked example: practising an interview answer

Speaking practice is where voice earns its place. Typing a polished answer to "Tell me about yourself" is easy. Saying it out loud, under a little pressure, is the skill you need on interview day.

Start the voice conversation and set up the role in your first spoken turn. Here is roughly what you might say:

```prompt
I have a job interview on Thursday for an office coordinator position at a community health center. I want to practise. Please act as the interviewer. Ask me one common interview question at a time and wait for my answer. After each answer, give me two pieces of feedback: one thing that worked and one thing to improve, especially if I rambled or forgot to give an example. Keep your feedback under thirty seconds. Then ask the next question. Start whenever you're ready.
```

A typical session then flows like this:

1. ChatGPT asks its first question, perhaps "Tell me about yourself."
2. You answer out loud, as you would in the room.
3. It gives short feedback, for example that your answer listed duties but no result, and suggests ending with one concrete achievement.
4. You say "Let me try that one again," and answer a second time.
5. After four or five questions, you say "Stop the interview and summarize my three biggest habits to fix."

When you exit, the transcript is in your chat history. Read it afterward: seeing your spoken answers in writing shows filler words and long detours more clearly than listening ever does.

> **Try it:** Pick a conversation you are nervous about, such as an interview, asking for a raise, or a presentation question period. Ask ChatGPT by voice to play the other person for five minutes, then read the transcript and note the one phrase you want to say better next time.

## Things to keep in mind

The spoken answers come from the same model as the typed ones, so they can contain the same mistakes, delivered in a friendly, confident voice. Check any fact that matters. Voice also tends to make replies feel more personal than they are; the feedback on your interview answer is a useful outside view, not an expert hiring opinion. And since conversations are hands-free once started, remember the session is running if you set the phone down.

## When is voice better than typing?

Voice is better when the skill you want is spoken, or when your hands and eyes are busy. Interview practice, rehearsing a speech, conversation practice in a language you are learning, talking through a recipe while you cook: in each case speaking is the point, or typing is impractical. OpenAI's own suggestions are in the same spirit, such as asking for a bedtime story or settling a debate at the dinner table.

Voice also helps when you are thinking out loud. Many people explain a messy problem more freely by talking than by writing, and a spoken back-and-forth can untangle an idea quickly.

Typing is still better when precision matters. A long prompt with exact figures, a document to analyze, or code to debug is easier to write, review and edit on a screen. Typed chats are also easier to skim and copy from afterward. A good habit is to mix the two: talk to explore and rehearse, then switch to text when you need a careful, checkable result.

## Sources

- Voice conversations (beta) - FAQ: https://help.openai.com/en/articles/8400625-voice-conversations-beta-faq
- ChatGPT can now see, hear, and speak: https://openai.com/index/chatgpt-can-now-see-hear-and-speak/
- ChatGPT — Release Notes: https://help.openai.com/en/articles/6825453-chatgpt-release-notes