Languages & Models

PreviousNext

Lanson Live speech capabilities and language settings for your integration.

Lanson Live

Lanson Live is LansonAI's speech model brand. Its model name is lanson-live-1, and you access its speech capabilities through Lanson Live API.

You do not need to choose a model to get started. Connect to the API, send audio, and configure the language and features your application needs. Trinity Engine handles the underlying processing; StableStream™ and LiveFinal™ describe its readable output and correction/finalization behavior.

Languages

Use a concrete language code when you know it:

  • zh — Chinese
  • en — English
  • auto — let the service auto-detect the language

Auto-detection is supported, but passing the known language is recommended because it is more accurate and avoids an extra detection step.

APIlanguage behavior
Realtime WSOptional. Omit it or send auto to auto-detect; a concrete code is recommended when known.
SegmentRequired. Missing or auto returns 400 invalid_language.
Batch jobsOptional. Omit it or send auto to auto-detect; a concrete code is recommended when known.

For the realtime gateway, update the language at any time with a session.update message. A concrete code always wins; sending "auto" (or null) resets detection and re-detects on later speech, so a session can start in English and switch to auto when another language appears.

Detecting the language first

If your client has a short audio sample and you want an explicit language before opening a realtime session, call POST /v1/audio/language-detection and pass the returned language to session.update or to the batch/segment request. Alternatively, omit the language (or send auto) and let the session's internal detection strategy decide.

Compatibility parameter

The Segment and Batch HTTP endpoints accept an optional model string for compatibility with existing clients. You can omit it. If a client requires a name, use lanson-live-1; this field does not select a different speech capability in the current implementation. Existing request values do not need to change.

The Realtime WebSocket API does not require a model parameter. Use session.update for language, refinement, and translation settings.

See also