How to Check Your English Pronunciation: 6 Methods and What Each One Really Tells You

Six ways to check your English pronunciation without a teacher: dictation, recording yourself, AI apps, phoneme scoring, minimal-pair listening and a human listener, with the blind spot in each.

Published on August 31, 2026

Direct answer

You can check your English pronunciation without a teacher, but no single method measures everything: dictation tests whether a machine recognises your words, recording tests what you actually produced, phoneme scoring tests which sound went wrong, and only a human tests whether you were understood. Use two or three of them together and each one covers the others’ blind spot.

1. The dictation test (free, and better than it sounds)

Open the dictation keyboard already built into your phone, say a sentence, and read what it typed. Every word it gets wrong is a word a listener may also have to work harder to recognise.

This works because speech recognition on iOS and Android does the same job a listener does: it turns your sounds into words. When it types tree for three or sink for think, it has found a real ambiguity in what you produced.

Two cautions. First, this is a rough instrument. Recognition systems are trained unevenly and misread accented English more often than native speech, so treat a wrong word as something to look into rather than a settled verdict. Second, dictation quietly uses context: say a word inside a predictable sentence and the system may correct it for you, hiding the problem. Test individual words as well as sentences.

2. Record yourself and listen back a day later

You hear the word you intended, not the word you produced, because you already know what you meant before you say it. Recording breaks that loop by letting you listen as a stranger would.

Record thirty seconds of something real: an answer you would give at work rather than a textbook paragraph. Then wait. Listening the next day, with the intended words no longer fresh, is far more revealing than listening immediately.

Listen once for each thing separately: once for word endings, once for stress, once for pace. Trying to hear everything at once produces a general feeling of dissatisfaction and no action.

3. Minimal-pair listening tests

Before you can fix a sound, you need to be able to hear it. Play ship / sheep, think / sink, bat / bet in random order and try to identify which is which without looking.

If you cannot hear the contrast reliably, no amount of repetition will fix the production, because you have no target to aim at. This is the cheapest diagnostic on the list and the one most people skip — minimal pairs in English covers how to choose the right contrasts for your first language.

Perceiving a difference does not mean you can produce it. Passing this test tells you the problem sits in your mouth rather than your ear, which is where to spend the next hour.

4. Dictionary audio: check the target, not yourself

If you are unsure of a specific word, a learner dictionary such as Cambridge gives you both the audio and the IPA transcription, usually in British and American versions. The British Council publishes similar reference material for learners.

This does not check your pronunciation. It checks your assumption about what the word should sound like, and a surprising share of “pronunciation problems” turn out to be a wrong mental model of the word, learned from spelling and never corrected. Fixing the target is faster than drilling the wrong one.

5. Pronunciation apps and AI checkers

They are consistent rather than authoritative: an automatic score tells you reliably whether today’s attempt resembles the model more than yesterday’s, but it cannot tell you whether a real person would have understood you.

This is the category that has grown fastest, so it helps to know what these tools actually compute. Most score your speech by comparing it against an acoustic model of the same words: some at sentence level, some sound by sound. The technical name for the sound-by-sound version is Goodness of Pronunciation; what GOP is and why a sentence score is not enough explains the method.

When you compare tools, four questions separate them:

• Does it tell you which sound failed, or only that something did? A single number per sentence is a verdict; a per-word or per-sound breakdown is an instruction. A sentence score tells you that something was off; phoneme-level feedback tells you which sound in which word was off, which is the difference between a verdict and an instruction.
• Can you practise your own text? Fixed lesson content will never contain your job title, your colleagues’ names, or the sentence you have to say on Thursday.
• Where is your audio processed? Voice is biometric data. On-device processing means the recording never leaves your phone; why on-device matters covers the trade-off.
• Is it an assistant or an assessor? A chatbot you talk to is not running a scorer at all; can ChatGPT or Gemini check your English pronunciation has a two-minute test that tells the two apart.
• Is there a daily limit? Pronunciation improves through repetition. A tool that caps attempts is pricing the exact behaviour that makes it work.

The blind spot applies to all of them, including good ones: the model is comparing you to an average of recorded speech. It cannot know that your listener is a colleague in Berlin who understands you perfectly well.

6. A human listener, once a month

Yes, occasionally. A person is the only check that measures whether a listener understood you the first time, so book one roughly once a month rather than every day.

Do not ask “how is my accent?”, because you will get politeness. Ask a specific, answerable question: “Read back what you heard”, or “Which word did you have to guess?” Those questions produce data. Everything on this list before it is a proxy for this one.

If you have no access to a teacher, a colleague, a language exchange partner, or anyone who will read back what they heard will do. The value is in the read-back, not in their qualifications.

Putting it together: a weekly check that takes ten minutes

1. Monday — dictation. Ten sentences from your real life. Write down every word it misheard.
2. Wednesday — recording. Thirty seconds, listened back on Thursday, one thing at a time.
3. Any day — targeted work. Take the two or three words that failed on Monday and practise those specifically, checking the dictionary audio first so you are aiming at the right target.
4. Monthly — a human. One read-back conversation.

Track words rather than scores. A list of words that used to fail and no longer do is evidence; a rising number is a feeling. How to measure English pronunciation progress sets out a fuller scorecard if you want to keep records properly.

How Phonema fits

Phonema covers method 5 and does it word by word: you say your own sentence, and the app marks which words came out clearly and which did not, all on your iPhone with no audio leaving the device and no daily cap. It does not replace the dictation test or the monthly human check. It sits between them, turning “something in that sentence was off” into a specific word you can practise again.

Bottom line

Six methods, three of them free, none of them complete. Dictation shows you what a machine failed to recognise, recording shows you what you actually said, minimal pairs show whether you can hear the difference, dictionary audio shows the target, an app shows which word broke, and a human shows whether it mattered. Run two of them this week and you will know more about your English pronunciation than a score alone can tell you.

Related sounds

• θ — as in think
• ɹ — as in red
• i — as in see
• ʌ — as in cup