Timed cues

Lyrics and subtitles by time: read LRC, SRT, WebVTT or Whisper JSON once, then ask for the line and the word at t.

1import { cueAt, loadTimedText, wordAt } from './type/timed-text.ts'23// in open4const lyrics = await loadTimedText(host, 'lyrics/song.lrc')  // .lrc, .srt, .vtt or .json; { trim } if the film starts later56// in render(t)7const line = cueAt(lyrics, t)    // { start, end, text, words? }, or undefined between cues8const word = wordAt(lyrics, t)   // { start, end, text }: the word being sung, when the file has words

It gives times only: how a line looks and moves is the film's code.

• LRC: [mm:ss.xx] line, several stamps on a line for a repeated line, enhanced LRC's <mm:ss.xx> before each word, and [offset:+250] (ms). A line ends where the next starts.
• SRT and WebVTT: start --> end blocks; tags such as <i> are stripped. No styling or positions are read.
• JSON: rows of [start, end, "line"], cues like the ones cueAt returns, or Whisper's segments with words (an aligner's output, for word times).
• parseLrc, parseSrt, parseTimedJson and parseTimedText(path, source) read text you already have; trimCues(cues, s) shifts every time by -s.