What problem does it solve?
After cutting a talking-head video, creating accurate subtitles normally requires re-transcribing the edited footage and manually fixing misrecognized proper nouns and sentence breaks. This Skill computes post-edit subtitle timing directly from the existing word-level transcript and the edit ledger, corrects misheard terms with a dictionary and the author's script, splits subtitles into per-sentence screens, and supports screen-by-screen human review in Studio.
Core Features & Use Cases
- Timing without re-transcription: Derives each word's position in the final cut by combining transcript.json timestamps with edit-list.json, producing subtitles.json with no extra ASR pass.
- Term correction with evidence: Applies a fixed dictionary for the speaker's standard spellings and aligns against the author's script only where context matches, reporting uncertain cases instead of guessing.
- Sentence-based screen splitting: Splits subtitles by punctuation and pause rules, merges one-word fragments, and reports stale, too-fast, or moved screens precisely.
- Use Case: After cutting filler words from a recorded talk, ask the agent to build subtitles; it reuses the existing transcript, fixes terms like product names via the dictionary, splits screens, and opens Studio so you can review each screen before export.
Quick Start
Use the chengfeng-subtitle skill to build subtitles for my edited project, fix the proper nouns with the dictionary, and open Studio so I can review each screen.
