sxsphinxstack

Skills / Working / Design a video caption workflow

Working Video & audio skill

Design a video caption workflow

Design a caption workflow for accurate text, timing, speaker identity, sound cues, localization, review, and delivery.

Treat captions as authored access content, not an unchecked speech-recognition export.

Procedure

  1. Define languages, formats, platforms, frame rates, accessibility standard, and owners.
  2. Establish the final audio and picture lock required for timing.
  3. Transcribe speech faithfully with names, terminology, punctuation, and meaningful sound cues.
  4. Segment lines for readable length, grammar, pace, and on-screen placement.
  5. Time captions to speech and relevant sounds without masking critical visuals.
  6. Distinguish speakers consistently and handle crosstalk, music, and inaudible content.
  7. Review by a qualified human, then proof in the actual player.
  8. Localize from an approved source transcript with cultural and timing adaptation.
  9. Deliver sidecar and burned variants with version, language, and QC evidence.

Guardrails

  • Never publish auto-captions without appropriate review.
  • Do not censor meaningful language unless the source itself is censored or policy requires notation.
  • Avoid placing captions over names, translations, or essential action.