SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsspace-video-subtitle

space-video-subtitle

Office & productivity

View repository

Updated 10-16 04:45

About this skill

space-video-subtitle · Subtitles

Subtitles aren't "done once transcription is finished". Spoken subtitles need to becomfortably segmented, with correct proper nouns, and aligned with the visuals. Four steps.

Workflow

视频/音频 → ① 转写(带时间) → ② 口播式断句 → ③ AI 校对 → ④ 输出 SRT/ASS →（可选）烧录

① Transcription

  • Extract 16k mono wav → ASR (any available backend: SenseVoice / whisper / Volc)
  • Produce time-stamped word-level or sentence-level results

② Spoken-style sentence segmentation

Subtitle segmentation is different from punctuation; segment bybreath and semantics, not by written punctuation:

  • One screen of subtitles ≤ one sentence readable at a glance (Chinese generally ≤15 characters/line, shorter for vertical screens)
  • Break at natural pauses, don't split a phrase across two screens
  • Filler words/colloquialisms can be kept (more authentic), or cleaned up per user preference

③ AI proofreading (most critical)

ASR first drafts always have errors; focus on correcting three types:

  • Proper nouns: names of people, product names, English abbreviations (Qoder, HyperFrames, MCP…) — build a glossary and feed it in for consistency
  • Homophones: highly frequent in Chinese ASR ("在/再", "的/得", "做/作")
  • Punctuation and numbers: "百分之八十七" in speech → "87%" normalized as needed

If there is editing (space-video-edit), subtitlesmust be re-transcribed and re-proofread based on the edited video, and the original video's subtitles cannot be reused.

④ Output

  • subtitles.srt: general
  • subtitles.ass: when styling is needed (font, outline, position, per-character karaoke), use
  • Naming corresponds to the finished video

Burning (optional)

# 软字幕(可关) —— 封装进容器
ffmpeg -i in.mp4 -i subtitles.srt -c copy -c:s mov_text out.mp4

# 硬字幕(烧进画面) —— 带样式
ffmpeg -i in.mp4 -vf "subtitles=subtitles.ass" -c:a copy out_hard.mp4

For vertical screens, mind the subtitle safe area (don't let platform UI block it); generally keep it above the bottom 1/5 of the frame.

Handoff

  • Upstream:space-video-edit 's source_cut.mp4, or any finished video/audio
  • Downstream: finished video with subtitles → space-video-cover or publish directly

Dependencies

ffmpeg, one ASR backend; burning ASS requires ffmpeg with libass.

Core principles

  • Segment by breath, not by punctuation
  • Build the glossary for proper nouns first: unify the glossary to avoid the same word being correct in half the episode and wrong in the other half
  • Re-transcribe after editing: for edited videos, subtitles must be redone