Skip to content

File transcription

Path: Settings → File transcription (same-named sidebar entry).

  1. Drop audio/video onto the drop zone (or pick a file).
  2. Uses the file-only ASR catalog entry (can differ from live voice typing).
  3. Copy the text when done; results may appear in history.

Open Choose model on the file page. Keys are shared with live voice typing; the catalog entry can be independent.

Runtime still uses isASRProviderAllowed:

  • Free: Apple / Doubao / four built-in on-device engines (SenseVoice, Whisper Turbo, Fun-ASR-Nano, Parakeet TDT)
  • Trial / Pro: full engine list

In demo mode the job may return fixed demo text instead of real ASR.

If voice typing / command / translation is recording, recognizing, polishing, or showing a result HUD, file transcription refuses to start (pipelineBusy). Finish or cancel the live session first.

Cloud file ASR is blocked — use Apple, a built-in on-device engine, or a loopback Base URL (plan gate still applies).