arabic-captions
v0.1.0
Published
Turn SRT/VTT subtitles into short keyword captions for Reels, TikTok and Shorts, with correct Arabic + English (bidi) rendering, Latin digits, filler removal and ASS output for ffmpeg. Zero dependencies.
Maintainers
Readme
arabic-captions
Turn any SRT or VTT file into short keyword captions for Reels, TikTok and Shorts: 2 to 4 words at a time, one highlighted word, Latin digits, fillers removed and Arabic + English words kept in the right order. Exports SRT, VTT, or ASS to burn in with ffmpeg. Zero dependencies.
حوّل ملف الترجمة لكابشن قصير بكلمات مفتاحية للريلز: كلمتين لأربعة كل مرة، كلمة وحدة ملوّنة، أرقام إنجليزية، بدون "يعني" و"اممم"، والعربي مع الإنجليزي بترتيب صح.

CLI
# 3 words per caption, as SRT
npx arabic-captions talk.srt -o talk.short.srt
# styled ASS → burn into the video with ffmpeg
npx arabic-captions talk.srt -f ass --font "Changa" --highlight "#7FB069" -o talk.ass
ffmpeg -i talk.mp4 -vf ass=talk.ass -c:a copy talk.captioned.mp4Get the SRT from any transcription tool (Whisper, CapCut, Premiere, YouTube Studio). Install the font on your machine first so ffmpeg can find it.
| Option | Default | |
| --- | --- | --- |
| -w, --words | 3 | words per caption |
| -f, --format | srt | srt, vtt or ass |
| --font / --size | Changa / 84 | ASS only |
| --color / --highlight | #FFFFFF / #7FB069 | ASS only |
| --keep-fillers | off | keep يعني / um / uh |
| --arabic-digits | off | keep ٠١٢٣ instead of 0123 |
| --no-bidi | off | skip direction isolates |
Library
import { parseSubtitles, keywordCaptions, toSrt, toAss, fixBidi } from "arabic-captions";
const cues = parseSubtitles(srtText);
const caps = keywordCaptions(cues, { maxWords: 3 });
// [{ start: 0, end: 980, text: "جرّبت سيارة BYD", words: [...], keyword: 2 }, ...]
toAss(caps, { font: "Changa", highlight: "#7FB069" });
fixBidi("جرّبت iPhone 17 لمدة 30 يوم"); // wraps English runs in Unicode isolatesHow the keyword is chosen: numbers first (they stop the scroll), then English words (brand and product names), then the longest word that is not a common filler or particle.
Timing: each chunk gets a share of the original cue's time based on its length, so captions stay in sync without a new transcription.
License
MIT © Eng. Moath Alherani
