Free English pronunciation checker
Pronunciation Test
Read this passage aloud
About 20 seconds · normal speaking paceLast Wednesday, my colleague, a photographer, asked me to buy three comfortable chairs and a sharp vegetable knife for her island cabin. I kept the receipt, although she thought the prices were a little high. Then we walked through the village in the rain, sat by the window, and watched the ship leave while the sheep slept on the hill.
Your recording is sent once to our speech-recognition provider and is not stored. 5 free tests a day, 20 when signed in.
Read a 20-second passage aloud and find out which words came across as something else. The test highlights each one in the text, shows what it sounded like — “sheep” heard as “ship”, “three” as “tree” — and explains how to say it. Your accent is never marked as a mistake. Free, no sign-up, and your recording is not stored.
How it works
Three steps, about a minute
- 1
Read the passage aloud
Press record and read it at your normal pace — about 20 seconds. No microphone? Upload a recording of yourself reading it.
- 2
See which words came across
In a few seconds the passage comes back with any word a listener would hear differently marked, what it sounded like, and how to say it.
- 3
Hear it said natively
Send the marked words to the generator below and hear them read by an American or British AI voice, as slowly as you like.
Examples
Practice passages, read by native AI voices
Every clip below was generated here on the free tier — no account, no credits. Play one to hear the voice on that kind of writing, or load the script into the generator and make your own version of it.
Silent letters
Wednesday, knife, island — letters you write but never say.
“On Wednesday I walked to the island with a knife, a receipt and an old comb. Listen closely: the half-hour talk about the castle has the answer in the last column.”
Word stress
PHO-to-graph, pho-TOG-ra-pher — one root, a moving stress.
“The photographer took a photograph for a photography magazine. Politics is political, the economy is economic, and a comfortable hotel is worth every penny.”
The -ed ending
Walked, played, wanted — three sounds for one spelling.
“She walked home, watched the news and asked for help. He played the guitar and called a friend. They wanted a break, so they waited, rested and started again.”
Included
What you get
- Word by word, in the text
- Marked words are highlighted in the passage itself, so you see exactly where a reading went off — not a vague overall impression.
- What it actually sounded like
- Each marked word shows what it was heard as — “bees” for “peas”, “keys” for “kids” — so the problem is concrete.
- A fix for every word
- Every word in the passage has a written tip on how to say it the standard way, where the stress falls and which letters stay silent.
- Your accent is not marked
- American and British pronunciations are both correct, and so is a steady accent that keeps every word clear. Only words that came across as something else are marked.
- Recording not stored
- The audio is converted in your browser, checked once, and dropped. Nothing is saved to your account or used for training.
- Free, no sign-up
- 5 tests a day as a guest and 20 with a free account. No app to install.
What this pronunciation test checks
The passage is short, but it is built so that one read walks you past the traps English learners fall into most — silent letters, syllables that should disappear, a word whose stress moves, th, long and short vowel pairs, and the three sounds of -ed. Every word you need to watch appears in one natural 20-second paragraph.
Your result marks only the words that came across as something else. The table shows what each trap looks like, so you can see why a word was marked and what the fix is.
| Trap | Where it is in the passage | The usual slip |
|---|---|---|
| Silent letters | Wednesday, knife, island, receipt, walked, high | Saying the letter that is written but not spoken — the k in “knife”, the p in “receipt”. |
| Swallowed syllables | comfortable, vegetable, colleague | Giving every written syllable its own beat: “com-for-ta-ble” instead of KUMF-tuh-bul. |
| Word stress | photographer | Stressing the first syllable, as in “photograph”, instead of pho-TOG-ra-pher. |
| th | three, then, although, thought, through | Replacing it with t, d, s or f — “tree”, “den”, “sought”, “free”. |
| Long and short vowels | ship / sheep, leave, hill, little | Merging the pair, so “sheep” becomes “ship” and “leave” becomes “live”. |
| The -ed ending | asked, walked, watched | Dropping it (“ask”) or adding a syllable (“walk-ed”). After k, ch or s it sounds like t. |
| v and w | village, window, while | Swapping the two, so “village” becomes “willage”. |
How it works — and why there is no score
A speech-recognition model writes down what it hears in your recording without being told what you were reading. That transcript is then lined up against the passage word by word. Where the two disagree — the model wrote “ship” where the text says “sheep”, or heard nothing where “sharp” should be — the word is marked, with what was heard.
Not telling the model the text is the important part. We first tried an AI model that listened with the passage in front of it, and it marked four to eight words on every perfect reading — the same famous “hard” words each time, whatever the audio said. A listener who does not know the script cannot be fooled that way: if it wrote down the word you meant, your pronunciation got it there.
That is also why you get marked words, not a percentage. “3 words to work on” is something you can check against the highlights; an overall score would look precise without meaning anything.
How accurate is it?
We tested it on readings where we knew the answer: clean readings by native-sounding AI voices, and readings where we planted specific mistakes. It never marked a word in a clean reading, and it caught every mistake that turned one word into another.
It is weaker on slips that leave a word recognisable: a pronounced silent letter or an extra syllable often still gets written down as the right word, so it goes unmarked. On real speech, we ran it over 469 words of native speakers reading audiobooks; almost all of the words it marked there were names, old-fashioned words, British spellings and numbers, none of which are in this passage — about four were ordinary words misheard. On recordings of learners from the Speech Accent Archive it picked out patterns teachers will recognise: “peas” heard as “bees”, “kids” as “keys”, “Stella” as “Estella”.
So read it this way: a marked word is one a listener could genuinely hear differently. An unmarked word came across, though a subtle difference may still be there — the practice passages above cover the ones this test is least likely to catch.
| What we tested | Result |
|---|---|
| Seven clean readings by native-sounding AI voices (American and British, three engines) | 0 words marked |
| Words swapped for another word — “sheep” read as “ship”, “chairs” as “cheers” | 4 of 4 caught |
| Words skipped or replaced | 3 of 3 caught |
| Consonant slips — “three” as “tree”, “thought” as “taught” | 3 of 6 caught |
| Silent letters and extra syllables — “canife”, “com-for-table” | 2 of 6 caught |
Pronunciation test or accent test?
They answer different questions. This test asks whether each word came across as the word you meant. The accent test asks which accent your English sounds closest to and lists the sounds behind it.
A steady accent is not a pronunciation problem — listeners adapt to it within minutes. What gets in the way is a sound that turns one word into another, and that is what this test marks. If you want to know how your English sounds overall, take the accent test as well.
Hear the passage read by a native voice
The generator below is loaded with the same passage. Play it in an American voice, switch to Emma for British English, and slow it down to hear exactly how each marked word is said. The button in your result loads just the words you need.
Speech studio
32 voices · 18 languages
What happens to your recording
Your browser converts the recording to a small mono audio file and sends it once to our speech-recognition provider, which returns a transcript. We do not store the audio or the transcript, attach either to your account, or use them to train anything. The only thing we keep is a daily count of tests per visitor, so the free limit works.
Questions
Frequently asked
Keep going