Skip to content
Skip to main content

New: Notetaker transcription language hints

NotetakerCalendar

Notetaker now accepts optional language hints when transcribing meetings. Use them when your users consistently speak one or more languages and Notetaker’s automatic language detection has been mis-identifying them — for example, Portuguese being transcribed as Spanish, or a single-language meeting bouncing between detected codes. Setting expected_languages either narrows what the auto-detector picks between (when you pass multiple codes) or forces a single language (when you pass one). Transcripts also now include a language field with the detected code, so downstream pipelines can branch on it without re-running detection.

This is a language declaration feature, not translation. Notetaker doesn’t translate transcripts or force code switching — the hints just constrain the candidate set so the recognizer doesn’t have to guess across every supported language.

Added

Section titled “Added”

  • meeting_settings.transcription_settings — Optional object with expected_languages (required when the object is non-empty; an array of one or more supported language codes) and fallback_language (optional; either auto or one of the codes already in expected_languages). To force a single language, pass one code in expected_languages. To narrow auto-detect, pass two or more. When fallback_language is omitted, Notetaker picks one from expected_languages based on detection confidence. transcription must be true for the hints to take effect.

  • Available wherever meeting_settings is accepted — Set on individual Notetakers (POST / PATCH, grant-based or standalone), on calendars (so all matching events inherit the setting), or on individual events (which override anything inherited from the calendar). Send transcription_settings: null or {} to clear inherited hints and return to default automatic language detection.

  • language field on transcripts — Notetaker’s transcript JSON now includes a top-level language field with the code of the language the recognizer detected. The field appears on both speaker_labelled and raw transcripts. See Handling Notetaker media files for sample payloads.

  • Set transcription languages guide — Full coverage of expected vs. fallback semantics, how clearing inherited hints works, and the complete list of supported codes.