How IPA Transcriber Works
From text to a checked transcript
- Your text is split into words, spaces and punctuation in this browser. Web addresses, email addresses and items with digits are kept exactly as written and marked for you to handle; numbers, dates and abbreviations are not expanded.
- Dictionary candidates. Each word is looked up in the pronunciation data of the system you chose. A word with one pronunciation gets it; a word with several (such as “read”) waits for your choice; a word that is not covered is marked “Not in this dictionary”.
- You check. You choose a pronunciation or type your own IPA for each occurrence separately. Your own IPA stays marked as your edit; it is never presented as dictionary data.
- You export the transcript as text, a table, a project file or a printed worksheet, always with its status and sources.
What it is not: it does not analyse recordings or anyone’s speech, it is not a clinical or diagnostic tool, and a transcript is not evidence of anyone’s learning. It does not choose between pronunciations from context. It produces broad transcriptions only, so there is no switch to “narrow” brackets.
Notation of each system
Both systems follow notation rules version conventions-v1. Only the symbol conventions listed below are changed from the source data; different vowels are never merged.
| Rule | General American | British RP-style |
|---|---|---|
| Level | Broad, dictionary reference pronunciation. | Broad, dictionary reference pronunciation. |
| R | Written r as this product’s phonemic symbol; it does not claim a trill. | Also written r. Linking and intrusive r are not added. |
| Vowels | DRESS ɛ, GOAT oʊ, FLEECE i, GOOSE u; the ɑ/ɔ difference is kept where the data has it. | DRESS e, GOAT əʊ, FLEECE iː, GOOSE uː; no American vowel conversion. |
| R-coloured vowels | Kept as the data writes them. This data writes every r-coloured vowel as ɝ and stressed STRUT as ə, so results contain neither ɚ nor ʌ. | No American r-coloured symbols. |
| Stress | Primary ˈ and secondary ˌ where the data marks them; never guessed. | Primary ˈ and secondary ˌ where the data marks them; never guessed. |
| Weak forms | Citation forms. A weak form appears only when the data lists it as a separate pronunciation to choose. | Citation forms. No automatic stress, linking or rhythm analysis. |
| Inside a word | No surrounding slashes in the stored IPA; syllable marks only where the data has them. | The same. |
| Narrow detail | No [narrow] option, aspiration, flaps or other phonetic detail. | The same. |
Symbol conventions applied to the data
| System | Written in the source | Written here | Reason |
|---|---|---|---|
| General American | ɹ | r | R notation: r is this product's phonemic symbol for /r/; no claim of a trill. |
| General American | ɫ | l | Broad vs narrow: no allophonic or acoustic detail; velarized [ɫ] is the /l/ phoneme in broad transcription. |
| British RP-style | ɹ | r | R notation: r is this product's phonemic symbol for /r/; no linking or intrusive r is generated. |
| British RP-style | ɫ | l | Broad vs narrow: no allophonic detail; velarized [ɫ] is the /l/ phoneme in broad transcription. |
| British RP-style | t͡ʃ | tʃ | Affricates are written tʃ, dʒ; the tie bar only marks the same affricate. |
| British RP-style | d͡ʒ | dʒ | Affricates are written tʃ, dʒ; the tie bar only marks the same affricate. |
| British RP-style | ɛː | eə | SQUARE written ɛː is the RP SQUARE eə. |
| British RP-style | ɛə | eə | SQUARE written ɛə is the RP SQUARE eə. |
| British RP-style | ɛ | e | DRESS written ɛ is the RP DRESS e. |
| British RP-style | ɪ̯ | ɪ | The non-syllabic mark in a diphthong is notation only. |
| British RP-style | ʊ̯ | ʊ | The non-syllabic mark in a diphthong is notation only. |
| British RP-style | ə̯ | ə | The non-syllabic mark in a diphthong is notation only. |
The symbols each system may use are listed in the English IPA chart. A pronunciation with any other symbol, or with bracketed phonetic detail, is held back from the data instead of being shortened.
Numbers, abbreviations and unknown words
Digits, dates, amounts and abbreviations are not read out automatically. Type the reading you need into the word’s own IPA box, or change the text to words and transcribe again. A word that is not in the data keeps its original spelling and is marked; nothing is predicted for it.
Evaluation, limits and versions
| System | Data version | Words in the data | Evaluation |
|---|---|---|---|
| General American | lx1-850f78dd69af | 125,927 | Not evaluated |
| British RP-style | Not available in this build. | ||
“Not evaluated” means no reviewed test set has been scored yet, so this site states no accuracy figure. When an evaluation exists it reports coverage as well as acceptance, with sample sizes and intervals, for each system separately.
Engine engine-1, notation conventions-v1, build b1-f4fa01e2c928. Changes are listed in the changelog and the data in data sources.
Known limits
- Words with several pronunciations always wait for your choice; there is no automatic choice from context.
- The General American data does not distinguish ʌ from ə or ɚ from ɝ (see the notation table above).
- Names, rare words, new words and misspellings are often not in the data.
- “Choices completed” means nothing is left to choose. It does not mean a transcript was checked by an expert.