For the complete documentation index, see llms.txt. This page is also available as Markdown.

Choosing an endpoint

Pick the endpoint that matches what you are scoring.

What you're scoring
Use
Notes

A single word, letter name, letter sound, or non-word

Use arpa_mark for letter sounds and non-words.

A phrase, sentence, or passage read aloud (you provide the text)

Add include_fluency, include_intonation, or markup_language as needed.

Spontaneous, open-ended speech (no reference text)

Add relevance_context for on-topic checks; detect_dialect / enforce_dialect for language detection.

A task-achievement activity (describe image, re-tell lecture, answer question)

Add task_question or task_context as needed.

Written text (essay, short, or chat writing)

Set include_feedback=1 for detailed feedback.

A raw phoneme sequence

Legacy — prefer score/word.

Whether text is in the Speechace lexicon (no audio)

Authoring-time check.

Last updated