Trust
Methodology & data sources
Speakur’s moat is a pronunciation graph — structured fields you can trust and correct — not a blog of thin pages.
Pronunciation methodology
- IPA first. When an upstream dictionary entry includes IPA, we render it server-side.
- Accent labels. US and UK play buttons use audio URLs that encode accent when available. A Native control appears only for a real non-US/UK source clip.
- Syllables & stress. Syllable counts come from Datamuse when available; stress descriptions follow IPA primary-stress marks (ˈ).
- Phonetic spelling. Respelling is derived from IPA or curated preferred forms for known hard words — labeled as best-effort when derived.
- Practice. Short word-specific cues replace cloned HowTo filler. HowTo JSON-LD may remain for search; visible copy must not be identical across the catalog.
Data sources
- Free Dictionary API — definitions, phonetics, and public-domain-style audio clips when present
- Datamuse — syllable counts and definition fallback when the primary API 404s
- Speakur curated notes — common mispronunciations for known hard words (
data/common-mistakes.json) - Browser speech — click-gated fallback when no dictionary clip exists
Audio attribution
Each word page lists Sources and audio provenance. Dictionary clips are preferred; TTS/browser speech is disclosed when used. We do not autoplay audio.
Last verified
Word pages carry a Last verified date set at generation or enrichment time. Regeneration or a successful data refill updates that stamp. Methodology last reviewed: .