Skip to main content

Recognition language

Set the recognition language via language.

  • For whisper, this is required.
  • For sommers, this is optional. Supported languages: ko (Korean), ja (Japanese). If omitted, the default is ko.

Supported languages​

Sommers model​

  • Supports Korean (ko) and Japanese (ja).
  • If language is not set, it defaults to ko (Korean).

Whisper model​

Language detection and multi-language (Whisper-only)​

In addition to language codes, language accepts two special values for language detection and multi-language processing:

1. detect - Single-language detection​

Automatically detects the language of audio that contains a single language. Use this when the entire audio is in one language but the language is unknown.

Examples:

  • The entire audio is in English.
  • The entire audio is in Korean.

2. multi - Multi-language processing​

Handles audio containing multiple languages and recognizes the language used in each sentence or segment.

Examples:

  • English, Korean, and Japanese sentences alternate in one audio file.
  • Different languages are used in different sentences.
  • International meeting recordings contain multiple languages.

language_candidates option (Whisper-only)​

Used with detect or multi to narrow candidates.

  • Default: ["ko", "ja", "zh", "en"]
caution
  • language is required for whisper.
  • For sommers, language is optional (supports ko, ja; default ko).
  • detect/multi and language_candidates are Whisper-only features (not supported by sommers).
  • Too many candidates may reduce accuracy.

Examples​

In the POST Request body, set language (required for whisper). Optionally provide language_candidates for detect/multi.

POST example: Example

Unified example (presets): Example