English to IPA

How to use the IPA translator

1. Type or paste up to 500 characters of English, a word, a sentence or a short paragraph, or tap one of the examples.
2. Read the line between slashes. ˈ stands before the syllable with the main stress, ˌ before a secondary one; punctuation stays where you typed it.
3. Check the table under the line: every word with all of its dictionary pronunciations, the stressed syllable and the syllable count. A word in ⟨angle brackets⟩ is not in the dictionary; its syllable count is an estimate from the spelling.
4. Press Listen (normal or slow) for the whole text, or the speaker next to a word for that word alone, and repeat after it.

How it works

Each word is looked up in the CMU Pronouncing Dictionary (Carnegie Mellon University), which records General American pronunciations in ARPAbet, a 39-symbol alphabet with a stress digit on every vowel. The tool ships the entries for the 100,000 most frequent English words (frequency ranking by wordfreq): 65,859 of them have a dictionary entry, 5,929 with more than one pronunciation. The lookup is case-insensitive, keeps apostrophes (don’t, o’clock) and transcribes a hyphenated word part by part when the whole is missing.

Every ARPAbet symbol is converted to IPA with the table below (ARPABET, Wikipedia). Two vowels depend on the stress digit, because the dictionary uses one symbol for both readings: AH is ʌ when stressed (hut /hʌt/) and ə when unstressed (about /əˈbaʊt/); ER is ɝ when stressed (hurt /hɝt/) and ɚ when unstressed (butter /ˈbʌtɚ/). There are no length marks: General American has no phonemic vowel length, so beat is /bit/, not /biːt/.

The syllable count is the number of vowel symbols in the pronunciation: comfortable has four (/ˈkʌmfɚtəbəl/) whatever its spelling suggests. To place the stress mark the tool divides the word into syllables by the maximal onset principle (English phonology, Wikipedia): consonants between two vowels go to the following syllable as far as they form a cluster an English word can begin with. In instead the cluster n-s-t splits as n | st, because st can start a word and nst cannot, which gives /ˌɪnˈstɛd/; in complete m-p-l splits as m | pl, which gives /kəmˈplit/. The mark goes before the whole onset, so banana is /bəˈnænə/ with ˈ before the n. One-syllable words carry no mark, as in dictionaries.

When the dictionary lists several pronunciations the line uses the first and the table shows all of them: record is /rəˈkɔrd/, /ˈrɛkɚd/ and /rɪˈkɔrd/, and the tool does not guess which one your sentence means. For a word outside the dictionary the syllable count comes from the spelling: count the vowel groups, drop a final silent e (make), keep a final -le after a consonant (table), and drop an -ed or -es ending that is not said as a syllable (walked, makes, but not wanted or boxes). The row says “estimate” when this rule was used.

ARPAbet to IPA reference table

The 39 symbols of the dictionary and the IPA the tool writes for each, with an example transcribed by the tool itself. AH and ER appear twice because their IPA depends on stress.

Symbols that are easy to confuse

What to do with the transcription

A sound you cannot produce (θ, ð, ɝ, æ, ʌ are the usual ones): work on it with the drills in the pronunciation practice guide. The 40 most difficult words to pronounce are a good test set.

Stress in the wrong place. If the transcription puts ˈ on a syllable you do not stress, play the word slowly and repeat it with the stressed syllable longer and louder; shadowing trains this at sentence level. For words whose stress changes with the meaning (record, present, object) see the words with multiple meanings guide.

A whole text. Transcribe a paragraph, listen once at normal speed, then read it aloud from the IPA line instead of the spelling; the English speaking practice texts are graded from A2 to C1 for exactly this.

Hear how you say it

The transcription shows the target. In the Fluently app the pronunciation drills score your English sounds one by one and the tutor corrects you while you talk. Installing the app and the spoken English assessment are free, on iOS and Android.

Download on theApp StoreGet it onGoogle Play

Limits: when this tool is not the answer

• General American only. Standard British forms (/ɒ/ in lot, /ɑː/ in bath, no r in car) are not shown; British spellings are looked up as typed and found only when the dictionary lists them.
• Citation forms, not connected speech. Each word is transcribed as said on its own. In a sentence weak forms (and as /ən/), linking and intonation change the sounds; the audio reflects that, the symbols do not.
• Meaning is not resolved. record, present, read and live show every pronunciation; pick the one that matches your meaning.
• Coverage. The dictionary was built for speech technology, not for learners: names, brands, new words and rare words may be missing or have a US-only reading. A missing word gets an estimated syllable count, no transcription.
• Numbers and symbols are not transcribed. Write 2026 as “twenty twenty-six”.
• Syllable boundaries follow one rule. Dictionaries sometimes keep a consonant with a preceding short vowel instead of moving it to the next syllable; the tool moves it whenever the result is a cluster English words can begin with, so in a few words ˈ sits one consonant earlier than in your dictionary. The syllable count is the same either way.
• The estimate for unknown words miscounts some spelling patterns: a y as the only vowel letter (as in rhythm), a vowel group said as two syllables (as in player) or an -ed said as a syllable after a consonant (as in naked) come out one syllable short. Those three words are in the dictionary; the rule applies only to words that are not.