Features

Languages

Set the audio language explicitly, list several expected languages, or let the API detect it automatically.

Three language modes are available on POST /v2/transcripts: automatic detection (the default), one explicit language via language_code, or a list of expected languages via language_codes for recordings that mix languages.

Automatic detection

Omit both language fields and the API detects the spoken language for you. The detected language appears on the completed transcript's language_code field.

Request body (auto-detect)

{
  "audio_url": "https://example.com/podcast.mp3"
}

Set the language explicitly

When you already know the language, set language_code with a lowercase code such as en, es, fr, de, pt, or it. An explicit language skips detection and is the reliable choice for short clips, heavy accents, or borderline audio quality.

Request body (explicit language)

{
  "audio_url": "https://example.com/entrevista.mp3",
  "language_code": "es"
}

Mixed-language audio

For recordings where speakers switch between languages, pass the expected set in language_codes. The transcriber uses the list as guidance while processing the audio.

Request body (mixed languages)

{
  "audio_url": "https://example.com/bilingual-call.mp3",
  "language_codes": ["en", "es"]
}

language_codes and language_code are alternatives: when language_codes is set it takes precedence and single-language handling is skipped.

Model coverage

Language coverage differs by model: whisper-voicekit-best concentrates on a set of core languages at the highest accuracy, while whisper-voicekit-fast covers a much broader range. If your audio is in a long-tail language, choose the fast model. GET /v2/models returns the current catalog and descriptions.