subtitle

Releases

1. ▸▾0.2.013 May 2026

   VTT and SRT parser rewritten as a line-based cue-block parser. The file is split on blank lines and only the --> timing line is matched with a regex,

   Performance

   • VTT and SRT parser rewritten as a line-based cue-block parser. The file is split on blank lines and only the --> timing line is matched with a regex, eliminating the catastrophic backtracking that caused multi-second hangs on large karaoke-style VTT files. (#21)
   • multiDurationSearch is now O(log n + k) — an upper-bound binary search locates the candidate window, then a backwards walk with a precomputed prefix-max-end array terminates as soon as no further active cue is possible.
   • Cue deduplication runs automatically after parsing and sorting. Same-range cues are collapsed to the longest text; contiguous cues with identical text and a gap ≤ 50 ms are merged into one. This keeps karaoke-style files manageable without losing visible content.

   Fixes

   • VTT cue identifiers (named cues such as intro or chapter-1) are now parsed correctly — cue names no longer bleed into .data. (#20)
   • VTT positioning directives (line:, position:, align:, etc.) on the timing line are silently skipped; only the cue text that follows is extracted.
   • Network and file fetching now attempt strict UTF-8 decoding first and fall back to Latin-1 for subtitle files served in legacy encodings. (#10)

   API additions

   • Subtitle.copyWith({index, data, start, end}) — creates a modified copy of a subtitle cue.

   Open source →