Subtitle Translation Workflow: Recognize, Translate, Re-attach
Subtitles are the first gate for going international. The good news: the whole chain runs locally in your browser — recognition produces timed captions, translation touches only the text, and publishing is a burn-in or sidecar decision.
Step one: generate source-language captions
Run local speech recognition to get an SRT: pick the Base model (about 78MB, most reliable for English and Chinese) and specify the language explicitly instead of auto-detect — explicit selection measurably reduces misheard filler words. After recognition, proofread once: proper nouns, brand names, and numbers are cheapest to fix at the text stage.
SRT is the universal interchange format: every entry is an index, a timecode line, and text. Any translation tool or caption editor understands it. Keep the original file untouched as your baseline and work on copies.
Step two: translate the text, never the timecodes
The iron rule of subtitle translation: timecode lines are untouchable. The format (00:00:01,000 --> 00:00:03,500) is rigid; one broken line invalidates the file. Extract the text entries, translate them line by line, and paste them back in order — feeding the whole raw file into a translator is how timecodes get mangled.
Two quality concerns: length and segmentation. One Chinese sentence often becomes a longer English one that overflows the frame; prefer idiomatic shortening over two-line wrapping. And segment by sense groups, not by mirroring the source punctuation — literal line breaks chop the rhythm.
Step three: re-attach and publish
For bilingual captions, source on top and translation below is the common layout. Publishing splits by platform: sites that accept caption uploads (YouTube and similar) work best with sidecar SRT/VTT — viewers can toggle and switch languages. For platforms where display must be guaranteed (messaging apps, embedded players), burn the subtitles into the frame with size and safe-area tuned for the target.
Convert between formats as needed: SRT to VTT for web players, to ASS for styled burn-ins. The entire pipeline stays on your device — unreleased footage never leaves the browser.