Speak Cards: 47 English Speaking Lessons for ChatGPT Voice Practice
Share a lesson with an AI voice tutor and work through an everyday situation in five to ten minutes
Tap the image to open the tool ↗The hard part of learning to speak a language is often not the knowledge. It is not daring to open your mouth, and having nobody to practise with. Real-time AI voice covers that gap, but open ChatGPT's voice mode with nothing prepared and many people run out of things to say within two minutes. I built Speak Cards (the site itself is titled 英語口說講義, English speaking handouts) to fill that gap. Each lesson is a handout written for the AI to read, so it can take you from vocabulary and sentence patterns all the way to a role-play, and you always know what to say next.
The lessons teach English, but the interface and explanations are in Traditional Chinese, so this guide gives the Chinese labels alongside their meaning.
What is in the handouts
There are 47 lessons live as of September 2026, with more on the way. They come in three levels, beginner A1, elementary A2 and intermediate B1, and each takes about 4 to 10 minutes. Every situation comes from daily life: ordering at a café, checking in at the airport, seeing a doctor, renting an apartment, missing the last train, getting separated from friends at a night market, the basics of a job interview and more.
Most lessons follow the full four-stage structure:
- Vocabulary: the core words and phrases for the situation, with American KK phonetic symbols, the Chinese meaning and an example sentence.
- Patterns: a few key sentence patterns that turn those words into sentences, each with examples.
- Roles: a model dialogue that goes back and forth, with an optional Chinese translation and a button to copy the whole dialogue.
- Scenario: a scenario card, such as a menu, a route map, a task card or a scene picture. The AI plays the server or the other person, and you talk freely using the information on the card.
A few shorter single-section lessons focus on one kind of content, such as a four minute set of kitchen vocabulary. The home page filters by level, practice focus (vocabulary, grammar, quiz, dialogue), lesson structure and topic tags, and has a search box for lessons or words. The English leans American.
Starting a lesson
- Pick a lesson. Choose a situation from the list on the home page and read the 學會這些 (you will learn) box at the top to see what the lesson covers.
- Open AI voice. Open ChatGPT on your phone or computer and start its advanced voice mode.
- Let the AI see the handout. Share your screen with the lesson page. If you cannot share your screen, press 複製整課提示詞 (copy the full lesson prompt) and paste the whole guide into the chat.
- Press 開始練習 (start practice). The page shows one stage at a time. Move between them with the previous and next buttons at the bottom, the left and right arrow keys, or by clicking 1 to 4. The full-screen button at the bottom right hides the site header and footer and enlarges the text, which makes the page easier for the AI to read over screen sharing.
- Finish with the scenario. After the role-play, ask the AI to review what you did well and what to work on.
Prompting the AI
The 給 AI (for the AI tutor) box on each lesson already contains a short English prompt. It asks the AI to coach you one stage at a time, wait for you to turn the page, speak slowly and correct your pronunciation and grammar as you go. You can switch the prompt level between A1, A2 and B1, and the instructions to the AI change to match. Each stage also has a green line of guidance for the AI, and its 完整指引 (full guide) button expands the detailed instructions for that stage.
If you prefer to brief the AI in your own words, start from something like this and adjust it to your level:
"You are my English speaking tutor. Use the vocabulary, patterns and dialogue on the screen to practise this lesson with me one stage at a time, and wait until I say next stage before moving on. Speak slowly, and when I make a mistake, correct my pronunciation and grammar gently and explain it in simple English."
If you are starting out, add a request to use simple words. If you want to practise reacting on the spot, ask the AI to throw in a follow-up question you have not prepared for during the role-play.
When you get stuck: rescue lines
The worst thing in speaking practice is getting stuck and stopping altogether. The 救援句 (rescue lines) button at the bottom right opens a small card of English lines you can read out when you are stuck, such as Sorry, can you say that again?, Can you speak more slowly? and I don't know how to say that yet., each with a Chinese translation. Getting comfortable with those lines is a useful speaking skill in itself.
Getting more out of it
- Shadow before you answer. Ask the AI to read a line and repeat it straight away, copying the rhythm and intonation, before moving into the conversation. For focused listening and shadowing, pair this with English Shadowing, which works one sentence at a time through YouTube videos.
- Ask for pronunciation correction explicitly. Tell the AI to correct your pronunciation, otherwise it will often just keep the conversation going.
- Swap roles. Once the dialogue feels comfortable, ask the AI to swap roles so you play the server or the other person.
- Little and often. One lesson of five to ten minutes a day is easier to keep up than a two hour session at the weekend.
Signing in and your data
Signing in is optional. You can read and practise every lesson without an account. After signing in with Google, a 標記為已練習 (mark as practised) button appears when you finish the last stage, and the lesson list and your account page remember which lessons you have done.
Signing in stores the name, email address (and whether Google has verified it) and profile picture URL from your Google account, your Google account identifier and the tokens Google issues at sign in, a record of each sign-in session (including the IP address and browser details you signed in from), and the lessons you marked as practised. Your voice conversation with the AI happens inside ChatGPT and is handled under OpenAI's privacy policy. Speak Cards never touches it. The site-wide details are in the privacy policy.
What it does not do
- It is not an AI. Speak Cards only provides the handouts. It does not listen to you or answer, so you need an AI voice service such as ChatGPT alongside it.
- Access to advanced voice depends on OpenAI's plans, so check OpenAI's own information for what your plan includes.
- Lessons currently go up to B1. There is nothing at B2 or above yet.
- The AI sometimes misreads the screen. If it drifts, telling it to go back to stage two on the screen usually brings it back.
Questions people ask
Do I need a paid ChatGPT plan? Availability and usage limits for advanced voice depend on OpenAI's current plans. The Speak Cards handouts themselves are free.
What level is this for? Lessons run from beginner A1 to intermediate B1, so they suit learners who want to speak but are not fluent yet. Stronger learners can use the scenarios to practise reacting on the spot.
Does the AI record what I say? The voice conversation runs inside ChatGPT and is handled under OpenAI's privacy policy. Speak Cards only provides the lesson material and never touches your conversation.
How does this compare with a human tutor? AI wins on availability, zero pressure and unlimited repetition. A human tutor wins on genuine interaction and on finer cultural and pragmatic corrections. Treat the AI as the place you speak every day and it does its job well.
To build listening and shadowing first, see English Shadowing. You can also try Spinning Top Arena, Emoji and Kaomoji, or go back to all tools.