fix(youtube): keep auto captions on screen for their full span

YouTube's sentence-level ASR emits long caption rows with a placeholder d="3000", while the caption actually displays until the next window event. Trusting `d` made long lines vanish mid-speech and leave a blank gap until the next cue. Rolling auto-caption documents (rows with a="1") now end each cue at the next event timestamp instead of t + d, matching YouTube's own display timing. Manual and non-rolling TimedText keep duration-based timing so real silence gaps are preserved.
2026-08-26 12:15:26 -07:00 · 2026-08-26 00:07:42 -07:00
parent c2c25c0da6
commit f8ca8681dc
3 changed files with 78 additions and 7 deletions
@@ -39,6 +39,36 @@ test('convertYoutubeTimedTextToVtt does not swallow text after zero-length overl
  );
 });

+test('convertYoutubeTimedTextToVtt extends rolling captions to the next window event', () => {
+  // Real-world shape of YouTube's sentence-level auto captions: window-append
+  // filler rows (a="1", sometimes without d) mark the display timeline, while
+  // long text rows carry a placeholder d="3000" far shorter than the speech.
+  const result = convertYoutubeTimedTextToVtt(
+    [
+      '<timedtext><body>',
+      '<p t="98550" d="3010" w="1" a="1">\n</p>',
+      '<p t="98560" d="3000" w="1"><s ac="0">ありがとうって言えないよね。こんなんじゃ。</s></p>',
+      '<p t="106950" w="1" a="1">\n</p>',
+      '<p t="106960" d="3799" w="1"><s ac="0">私だったら無理だよ。</s></p>',
+      '</body></timedtext>',
+    ].join('\n'),
+  );
+
+  assert.equal(
+    result,
+    [
+      'WEBVTT',
+      '',
+      '00:01:38.560 --> 00:01:46.950',
+      'ありがとうって言えないよね。こんなんじゃ。',
+      '',
+      '00:01:46.960 --> 00:01:50.759',
+      '私だったら無理だよ。',
+      '',
+    ].join('\n'),
+  );
+});
+
 test('normalizeYoutubeAutoVtt strips cumulative rolling-caption prefixes', () => {
  const result = normalizeYoutubeAutoVtt(
    [