Interactive learning material

The ChatGPT voice module

Spoken dialogue with artificial intelligence in teaching and learning

Author: Zsolt LengyeliTStudy Hungary Training and Research Centre
Reviewer: Mária HartyányiiTStudy Hungary Training and Research Centre

Sound waves travelling in both directions between the user and the AI — Talking with AI: spoken questions and answers

Learning objective

Talking is the most natural way of communicating – and AI assistants can now hold spoken conversations in real time. The aim of this material is to introduce voice-based AI interaction and its possible role in teaching and learning. By the end of the material, teachers will be able to distinguish dictation from voice conversation, describe how the current ChatGPT Voice works, identify meaningful uses for brainstorming, language practice, accessibility, quick professional assistance and reflective dialogue, and take into account practical limitations such as transcription errors, classroom noise and data protection when using voice with learners.

Introduction

Dictation or voice conversation?

Dictation

ChatGPT can be used by voice in two practical forms. Dictation simply converts our speech into text: we speak, the system transcribes, and we send the message as if we had typed it. This is useful when we want to formulate longer thoughts quickly or typing is inconvenient, for example on a phone.

Speech → text → sent message

Voice conversation

Voice conversation is a real dialogue: we speak to the AI, and it answers aloud. This allows a faster, more natural interaction – especially for brainstorming, practising or when we do not want to type continuously. The key difference: dictation turns speech into text, while voice conversation is a conversational interface in which the AI’s answer also arrives as speech.

Speech ⇄ spoken AI response

How voice works

How does ChatGPT Voice work?

Voice has developed quickly. Since late 2025 voice conversations take place directly inside the chat, so the spoken exchange and any visual content (for example a map, a table or an image) appear in the same conversation. In July 2026 OpenAI introduced GPT-Live, a new generation of voice models that can listen and speak at the same time. This makes turn-taking and interruptions feel natural: we can cut in, the model can signal that it is listening, and the conversation flows more like a human dialogue.

Full-duplex operation means that the user and the AI can speak and listen at the same time. We can clarify our request while the conversation continues.

Two-way voice communication between the user and the AI — Full-duplex voice conversation: the user and the AI can speak and listen at the same time

Live mode

Live mode (the default for most users) supports natural back-and-forth conversation and can use web search and memory; for complex reasoning or research it can hand over the task to a stronger text model while the conversation continues.

Advanced mode

Advanced mode keeps earlier real-time capabilities, including sharing the camera or screen on mobile devices – useful for showing an object, a device or a problem.

How voice works

There is still work to do after the conversation

Project context

Voice can also be used within Projects, so the spoken conversation can rely on the project’s files and instructions.

Transcript

After the conversation, a transcript appears in the chat history. It may not exactly match what was said, especially with overlapping speech, background noise or fast conversation.

Language and usage limits

Sessions have usage limits that depend on the subscription. The language of speech recognition can be set in the settings, which improves accuracy.

Check the transcript too!

The written version helps you develop your ideas further, but it is not a verbatim record. Review specialist terms, names, numbers and any details you will rely on later.

Educational applications

From ideas to practice

Brainstorming and planning on the go

Many teachers have their best ideas while commuting or walking. A spoken conversation is a quick way to think aloud about a lesson: “I teach tyre changing to first-year apprentices tomorrow. Help me find a motivating opening.” Afterwards, ask for a written summary in the chat and edit it.

Language practice

Voice is especially valuable for practising professional communication in a foreign language: a receptionist greeting a hotel guest, a mechanic explaining a repair to a customer, a waiter taking an order. The AI can play the role of the partner, speak more slowly or faster, repeat, and give feedback on vocabulary. Feedback on pronunciation should be treated with caution; the teacher remains the reference.

Four panels showing symbols for brainstorming, conversation, accessibility and reflection — Brainstorming, language practice, accessibility and reflective dialogue – four typical uses of voice interaction

Educational applications

Support without reading or typing

Accessibility

Voice interaction can lower barriers for learners with reading difficulties, visual impairment or motor difficulties: they can ask questions and receive explanations without reading or typing. For these learners, voice can be a genuine support – within the school’s framework for assistive technology.

Quick professional assistance

In workshops and practical settings, hands are often busy. A spoken question (“What torque is usually specified for…?”) can be faster than typing. However, safety-critical information must always be checked in the manufacturer’s documentation or the relevant standard – a confident spoken answer is not more reliable than a written one.

Professional verification is the teacher’s responsibility

For tightening torque, workplace safety requirements or other safety information, use the manufacturer’s documentation and the relevant standard as your reference. A voice response can be a starting point, but it does not replace verification.

Educational applications

A conversation partner for reflection

Voice conversation is well suited to reflection: talking through a lesson that did not go as planned, preparing for a difficult conversation with parents, or – for learners – practising a job interview. Asking the AI to take the role of a critical friend who asks questions, rather than one who simply agrees, makes the reflection more useful.

“Help me reflect on today’s lesson. Do not just agree: ask about my decisions and help me think through other approaches. Ask one question at a time.”

The role of a “critical friend”

The aim is to clarify our own thinking. A useful conversation offers questions, alternatives and new perspectives as well as encouragement.

Learners can practise a job interview, and teachers can rehearse a difficult conversation with parents, without sharing personal data.

Practical limitations

A quieter setting, a more accurate conversation

Transcription errors: accents, dialects, mixed languages and specialised terminology can be misrecognised. Check the transcript when the details matter.

Noise: background noise and several people talking at once can trigger unintended responses. Voice works best one-to-one; in class, use a headset or a quiet corner, and on iPhone the voice isolation setting can help.

A noisy group on the left and an individual using a headset on the right — In a noisy classroom, recognition suffers; voice works best one-to-one in a quiet setting

Before using voice in class

Choose a quiet corner, set the speech recognition language and agree that only one person will speak at a time. Check the transcript when important details are involved.

Practical limitations

A fluent answer can still be wrong

Ask for the sources!

No visible sources: a spoken answer does not show its sources as clearly as a written one. For facts, ask for the sources in the chat and check them.

Be aware of persuasive delivery!

Persuasiveness: a fluent, friendly voice can make wrong answers sound more convincing. Hallucinations do not disappear in voice mode.

Allow for the limits!

Limits: session length and usage limits apply, and some functions (e.g. screen sharing) are available only in certain modes.

Data protection

Data protection and appropriate use

Audio clips and the transcript of voice conversations are linked to the chat history. The provider describes a 30-day retention period for audio clips in Live and Advanced modes; this does not mean that the transcript is automatically deleted.

Audio clips are not used for training by default, unless you choose to share them. The use of transcripts depends on separate data settings and your subscription. Check the current terms and settings before use.

Details: OpenAI – ChatGPT Voice, data controls.

A microphone, a transcript, a shield with a padlock and a clock — Voice recordings and transcripts are personal data: know how they are stored and who can access them

Voice is personal data

A person’s voice is personal data. Do not record learners or classmates without consent, and follow the GDPR and the school’s data protection rules.

Age requirements and school rules

Respect the providers’ age requirements (for ChatGPT: at least 13 years, and parental consent under 18) and the school’s rules. With younger learners, use teacher-led demonstrations.

Agreed rules for use

Agree on rules: no personal or sensitive information in voice conversations, and a clear purpose for each use.

Takeaways

Takeaways for teachers

Five points to take away

Distinguish dictation (speech to text) from voice conversation (spoken dialogue).
Voice is powerful for brainstorming, language practice, accessibility and reflection.
Set the language, use a quiet setting, and check the transcript when details matter.
Spoken answers are not more reliable than written ones: verify facts and safety information.
Treat voice as personal data: consent, age requirements and school rules apply.

Start with one clear purpose!

Choose a specific situation: lesson brainstorming, professional language practice, an accessible explanation or reflective dialogue.

For your first attempt, choose a quiet setting and an example without personal data. At the end, ask for a written summary, then edit and check it.

You can download the accompanying practical task from the Credits page.

Self-check quiz

Check your knowledge!

Each question has one correct answer. After submitting your answers, you will receive a short explanation.

1What is the key difference between dictation and voice conversation?With dictation, the AI always answers aloud.Dictation turns speech into text; in voice conversation, the AI also responds aloud.Dictation can only be used for language practice.

2What does full-duplex operation mean?The user and the AI can speak and listen at the same time.The conversation can only take place in two languages.Every answer is produced in two copies.

3How should AI feedback on pronunciation be treated?Always accept it as a teacher’s assessment.It completely replaces the teacher’s judgement.Treat it with caution; the teacher remains the reference.

4Which setting is best for using voice?A noisy classroom where several people speak at once.A quiet setting, one-to-one dialogue and a checked transcript when needed.Fast speech and continuous language switching without checking.

5What should you do after receiving a safety-critical professional answer?Check it in the manufacturer’s documentation or the relevant standard.Accept it if the AI sounds confident.Ask the AI to repeat it more loudly.

6How should we prepare to use voice-based AI with learners?A teacher-led lesson replaces the need for learners’ consent.Voice recordings do not need to be treated as personal data.Check age requirements and school rules; avoid personal and sensitive data.

Sources and attachments

Credits and copyright

Author: Zsolt LengyeliTStudy Hungary Training and Research Centre
Reviewer: Mária HartyányiiTStudy Hungary Training and Research Centre
Project identifier: M1-T2-MLC3 · VITA

Attachments

Source material · PDFM1-T2-MLC3-TEXT-EN.pdf Practical task · PDFM1-T2-MLC3-EXC-EN.pdf

This work may be reused and adapted under the terms of the Creative Commons Attribution-ShareAlike 4.0 International License (CC BY-SA 4.0).

Funded by the European Union. Views and opinions expressed are those of the author(s) only and do not necessarily reflect those of the European Union or the National Agency.

References and further reading

OpenAI (2026): Introducing GPT-Live. 8 July 2026.
OpenAI Help Center (2026): ChatGPT Voice – FAQ; ChatGPT release notes.
Lengyel, Zs. (2026): Concrete applications for content generation. VITA training presentation.