# Pronunciation data, notation and licenses

PronounceKit independently converts open dictionary data to American dictionary
respelling and British broad IPA. These use Oxford-style notation conventions,
but are **not Oxford or Google dictionary data**. There is no affiliation,
endorsement, or guarantee of matching any particular dictionary edition.
No Oxford entries, recordings, definitions or examples are downloaded or bundled.

## One reference reading

Each word has at most one displayed reading **per accent**. An editorial choice
in `src/data/pronunciation-corrections.json` takes priority; otherwise the first
reading in the pinned source is used. This is a deterministic reference policy,
not a frequency ranking. `read` uses its present-tense reading, `wind` its
air-movement noun, `a` its weak form, and `enter` retains /t/.

There is no sentence analysis, homograph disambiguation or connected-speech
transcription. Other readings remain valid. Browser/system speech may select
a different reading. Missing words and missing accent entries stay unavailable;
one accent is never generated from the other.

## Display rules

The project-authored `src/lib/phonetics/notation.ts` applies the same rules to
dictionary entries, the initial homepage example, word lessons and letter names.

- American dictionary respelling: `ber`, `həˈlō`, `SHip`, `CHip`, `THiNGk`.
  `ā ē ī ō` represent /eɪ i aɪ oʊ/; `a ä ô` represent /æ ɑ ɔ/;
  `o͞o o͝o` represent /u ʊ/; `ou oi` represent /aʊ ɔɪ/.
  `ə` represents both weak schwa and the STRUT vowel in this convention.
  `ər` represents a rhotic vowel. `SH ZH CH j TH T͟H NG y` represent
  /ʃ ʒ tʃ dʒ θ ð ŋ j/. Remaining ordinary consonants retain their letter.
  Source /iɹ/ and /ɪɹ/ use `ir` for NEAR. Explicit editorial boundaries
  preserve /i.ɹ/ across prefixes, as in `prerecord` and `reregulation`.
- British broad IPA: `bear` is /beə/. DRESS is /e/, the approximant is /r/,
  dark-l detail is omitted, and source /ɐ/ is represented by /ə/.
  Stressed source /iə/ becomes /ɪə/ for NEAR; unstressed /i.ə/ in
  `area`, `alias` and `academia` remains two vowels.
  British word-final linking r is not added to isolated reference forms.
- Single-syllable words omit stress. For source stress marks placed directly
  before a vowel, a legal English onset is moved after the mark. This is a
  reproducible syllable-onset heuristic, not lexical or specialist review.
  Primary/secondary stress and accent-specific vowel choices otherwise remain.
- Slashes delimit readings in the UI; they do not turn American respelling into
  IPA. The localized labels and symbol key identify the system. IPA lessons,
  phoneme charts and IPA recognition exercises explicitly teach IPA itself.

The symbol conventions can be compared with the
[Oxford pronunciation guide](https://www.oxfordlearnersdictionaries.com/us/about/pronunciation_english.html)
and the New Oxford American Dictionary's pronunciation key.
These are references about notation, not licensed sources of our word database.
The full source corpus has not received pronunciation-specialist review.

## Sources and redistribution

The fixed data source is
[open-dict-data/ipa-dict at 43c3570](https://github.com/open-dict-data/ipa-dict/tree/43c3570eb3553bdd19fccd2bd0091534889af023),
retrieved 2026-09-20. Its English files and upstream notices are vendored in
`third_party/ipa-dict/`. Additional upstream license notices were retrieved on
2026-10-20. Each retains its own license:

- US: ipa-dict and cmudict-ipa, MIT; CMUdict's permissive notice is also retained.
  Upstream used Kyle Gorman's MIT-licensed syllabify to add stress.
- UK: derived from Melissa Boiko's ipacards, GPL-3.0. The UK derivative data
  remains GPL-3.0; commercial use does not remove its redistribution obligations.
- The mixed homepage example JSON is distributed under GPL-3.0, with the
  US notices retained. The build and notation code and project-authored
  corrections in this source package are also available under GPL-3.0.
  No claim is made that the whole mixed data package is MIT-only.

Download the readable data, conversion sources and licenses from this site:

- [US source](/phonetics/sources/en_US.txt) · [UK source](/phonetics/sources/en_UK.txt)
- [Upstream provenance](/phonetics/sources/UPSTREAM-README.md)
- [IPA Dictionary MIT](/phonetics/sources/ipa-dict-MIT.txt) · [cmudict-ipa MIT](/phonetics/sources/cmudict-ipa-MIT.txt)
- [CMUdict notice](/phonetics/sources/CMUDICT-LICENSE.txt) · [syllabify source and MIT notice](/phonetics/sources/syllabify-UPSTREAM.py.txt)
- [GPL-3.0](/phonetics/sources/GPL-3.0.txt)
- [Editorial choices](/phonetics/sources/pronunciation-corrections.json)
- [Notation conversion](/phonetics/sources/notation.ts.txt) · [Build script](/phonetics/sources/build-phonetic-dictionary.mjs.txt)
- [Homepage example data](/phonetics/sources/example-pronunciations.json)

The complete repository and reproduction environment are at
[Guitarja/pronouncekit](https://github.com/Guitarja/pronouncekit).
Use its `.node-version`, run `npm ci`, then `npm run build:phonetics`.
No network requests or paid API credentials are needed during generation.
The build replaces only `public/phonetics/` and the generated homepage example.
The manifest records source, editorial and notation-rule SHA-256 checksums;
all three participate in browser cache invalidation. Generated changes date from
2026-10-20. Git attributes keep these text inputs and downloadable sources in LF
format on all platforms. Rebuilding unchanged inputs produces identical data.

Existing editorial corrections cite sources for individual pronunciation facts;
they do not copy dictionary prose or claim exclusive ownership of those facts.
See [Oxford's API terms](https://developer.oxforddictionaries.com/api-terms-and-conditions?tab=commercial)
for why their licensed content is not used in this freely distributed static dataset.