How Speechace scoring works

Speechace scores spoken (and written) language through a small set of endpoints. Which one you use comes down to two things: whether you provide the expected text, and what you want scored.

Scripted vs. spontaneous

Scripted — you know what the speaker should say and pass it as the reference:

• score/word — a single word, letter name, letter sound, or non-word.

• score/text — a word, sentence, or passage read aloud; adds fluency, lexical stress, and intonation on request.

Spontaneous — there is no reference text, so the API transcribes what was actually said and scores that:

• score/speech — open-ended speech: pronunciation, fluency, grammar, vocabulary, and coherence, plus optional relevance and language detection.

• score/task — task achievement: how completely a response fulfills a specific task (describe an image, re-tell a lecture, answer a question).

Writing

• score/writing — essays and short written responses.

What each endpoint scores

Endpoint	You give it	It returns
score/word	One word / letter / non-word + audio	Word, syllable, and phoneme quality scores
score/text	Reference text + audio	Pronunciation — plus fluency, stress, and intonation on request
score/speech	Audio (no reference text)	Transcript + pronunciation, fluency, grammar, vocabulary, coherence (+ relevance, language detection)
score/task	Task prompt + audio or text	Task-achievement score, plus the language scores above
score/writing	Prompt + written answer	Task response, grammar, vocabulary, coherence

Every response is mapped to the common scoring rubrics — Speechace, IELTS, PTE, TOEIC, and CEFR — so you can compare levels across activities and tests.

Not sure which one you need? See Choosing an endpoint.