> ## Documentation Index
> Fetch the complete documentation index at: https://docs.langcraft.world/llms.txt
> Use this file to discover all available pages before exploring further.

# Langcraft Speech API

Add human‑level pronunciation assessment to your app with a single API call. Built for EdTech, speech therapy, and linguistic analysis.

<CardGroup cols={2}>
  <Card title="Getting started" icon="rocket" href="/quickstart">
    Make your first API request in minutes.
  </Card>

  <Card title="Output reference" icon="brackets-curly" href="/output">
    Learn the JSON schema and fields.
  </Card>
</CardGroup>

## API overview

Langcraft Speech API provides pronunciation assessment and prosody analysis in a single call. It’s designed for EdTech, speech therapy, and linguistic analysis, and returns structured JSON you can use directly in your app.

Key capabilities:

* Word and phoneme alignment with millisecond timing
* Phoneme‑level scoring and error detection
* Automatic transcription with word‑level timestamps when no reference text is provided
* Multilingual support (40+ languages)
* Model selection: send `model=aurora-1` for multilingual adult speech, `model=nova-1` for early-childhood speech, or leave `model` unset for standard English analysis

## Highlights

* Per‑phoneme scores with timestamps
* Per‑word rollups and summaries
* Pitch and stress contours at the phoneme and word levels
* Alignment metadata to connect audio, phones, and text

## Inputs

You can analyze speech with any of:

* A reference text (`reference_text`) plus language code (`lang`) — the API runs grapheme‑to‑phoneme generation to derive canonical phones
* A direct IPA phone sequence (`reference_phones`) — bypasses G2P, useful for pronunciation contrast tests
* Audio only — the API runs automatic transcription and uses the transcript as the reference

`reference_text` accepts the alias `text`. `reference_phones` accepts the alias `ipa`.

## Model selection

For non-English speech, set the official public model selector:

```text theme={null}
model=aurora-1
```

`aurora-1` is an experimental public model selector designed for languages other than English. It is currently recommended for German, French, and Spanish.

For early-childhood speech, use:

```text theme={null}
model=nova-1
```

`nova-1` is designed for speakers from approximately six months to eight years
old. See [Early-childhood speech analysis](/early-childhood-speaker-profile)
for supported languages and request requirements.