Files
hyperframes/packages/cli/src/docs/transcribe.md
T

1.5 KiB

Transcribe

Word-level timestamps for captions, or an import of an existing transcript.

Run it

  • hyperframes transcribe audio.mp3 --model small.en — known English
  • hyperframes transcribe video.mp4 --model small --language es — known non-English; no .en model
  • hyperframes transcribe subtitles.srt — import an SRT, a VTT or an OpenAI Whisper JSON response
  • hyperframes transcribe transcript.json --to srt — export captions; --to vtt for VTT, --output for the path
  • --json prints the result as JSON and sends progress lines to stderr

Pick the model

  • Always pass --model. The default small.en translates non-English audio into English.
  • .en models (tiny.en, base.en, small.en, medium.en) are English only. Use small, medium or large-v3 with --language for anything else.
  • Speech over music: start with medium.en.

Pick the engine

  • --engine auto (default) uses Parakeet when it is installed and covers --language, otherwise whisper
  • --engine whisper or --engine parakeet force one; install Parakeet with hyperframes models install parakeet
  • --engine heygen uses HeyGen's speech model; it needs an API key and is charged to API credits. --speakers labels each word with its speaker
  • --no-runtime-install refuses to install the whisper runtime; --optional exits 0 when whisper is unavailable
  • --timeout <ms> raises the whisper time limit on slow machines (minimum 5000)

Run hyperframes transcribe --help for every flag.