fix(subtitles): merge wrapped positioned caption rows (#234)

changes/caption-row-sentence-join.md

@@ -0,0 +1,4 @@
+type: fixed
+area: overlay
+
+- Broadcast-style Japanese caption tracks (Crunchyroll JA subs) that split one sentence across two positioned events now publish it as a single line, so `preserveLineBreaks: false` flattens it, the sidebar lists it once, and mined sentences are whole. Rows from two different speakers, sound effects, and labeled turns still stay on separate lines.

docs/architecture/subtitle-overlay-priming.md

@@ -134,6 +134,16 @@ coming and prefetching would otherwise idle for the rest of the cue.
   mpv's raw live text can be reconciled without displaying or mining the reading. The
   timing tracker (clipboard copy, recent-line mining) and immersion recorders run the same
   reconciliation on the `sub-start`/`sub-end` sample, so they record what the overlay shows.
+- Broadcast-caption rows that spell one utterance across several same-timed positioned events
+  (same style, layer, and vertical band, stacked at most two text rows apart) are joined into
+  one cue with a single line break, so `preserveLineBreaks` treats them like an authored `\N`,
+  and the recorders above see the whole sentence. A row continues the one above it when that row
+  is a bare speaker label, ends without terminal punctuation, or leaves a ≪…≫ / ⸨…⸩ span open; a
+  lower row that opens its own label or span always starts a new cue, which keeps two speakers
+  sharing the screen on separate lines. The pass runs only on scripts that read as broadcast
+  captions (a meaningful share of events carry speaker labels or ≪…≫ / ⸨…⸩ spans) and only on
+  rows containing Japanese, because fansub typesetting stacks positioned rows for signs, chat
+  bubbles, and headlines where that punctuation convention does not hold.
 - Fragment-only ASS karaoke is reconstructed per style before publication. Explicit spaces
   survive concatenation. Latin fragment typesetting with no literal spaces also recovers word
   boundaries represented only by materially larger horizontal `\pos` or `\move` gaps within that

src/core/services/subtitle-cue-parser.test.ts

@@ -1449,15 +1449,17 @@ test('parseSubtitleCues keeps tall CC-style base dialogue publishable after remo
     'Dialogue: 0,0:00:06.11,0:00:10.11,Default,,0,0,0,,{\\pos(172,437)\\fscx50}（{\\fscx100}立希{\\fscx50}）',
     'Dialogue: 0,0:00:06.11,0:00:10.11,Default,,0,0,0,,{\\pos(332,443)\\fscx50\\fscy50}ともり',
     'Dialogue: 0,0:00:06.11,0:00:10.11,Default,,0,0,0,,{\\pos(192,497)}お前…{\\fscx50}　{\\fscx100}燈をバンドに誘ったの？',
+    // A second labeled turn, so the script reads as broadcast captions.
+    'Dialogue: 0,0:00:10.11,0:00:12.00,Default,,0,0,0,,{\\pos(192,497)\\fscx50}（{\\fscx100}燈{\\fscx50}）{\\fscx100}うん。',
   ].join('\n');
 
   const cues = parseSubtitleCues(content, 'test.ass');
+  // The bare speaker label row joins the dialogue row beneath it as one cue.
   assert.deepEqual(
     cues.map((cue) => cue.text),
-    ['（立希）', 'お前…　燈をバンドに誘ったの？'],
+    ['（立希）\nお前…　燈をバンドに誘ったの？', '（燈）うん。'],
   );
-  assert.deepEqual(cues[0]?.assFurigana, ['たき']);
+  assert.deepEqual(cues[0]?.assFurigana, ['たき', 'ともり']);
-  assert.deepEqual(cues[1]?.assFurigana, ['ともり']);
   assert.ok(cues.every((cue) => cue.assLayout?.kind === 'positioned'));
 });
 
@@ -1485,14 +1487,184 @@ test('parseSubtitleCues removes half-size positioned furigana from broadcast cap
     [
       '（山田）ごめん　結局　ぬれたな。',
       '大丈夫。',
-      '（山田の母）ほんなら',
+      '（山田の母）ほんなら\n隠し貯蔵のミルクまんじゅう➡',
-      '隠し貯蔵のミルクまんじゅう➡',
       '絶対違う',
     ],
   );
   assert.deepEqual(cues[1]?.assFurigana, ['だいじょうぶ']);
-  assert.deepEqual(cues[3]?.assFurigana, ['かく', 'ちょぞう']);
+  assert.deepEqual(cues[2]?.assFurigana, ['かく', 'ちょぞう']);
-  assert.deepEqual(cues[4]?.assFurigana, ['ぜったい　ちが']);
+  assert.deepEqual(cues[3]?.assFurigana, ['ぜったい　ちが']);
+});
+
+// Broadcast-caption rows from You and I Are Polar Opposites S02E09. Every pair shares
+// timing, style, and the bottom band; only the text tells a wrap from a second speaker.
+const captionRowsHeader = ['[Script Info]', 'PlayResY: 540', '', ...eventsHeader];
+
+function captionRow(start: string, end: string, x: number, y: number, text: string): string {
+  return `Dialogue: 0,${start},${end},Default,,0,0,0,,{\\pos(${x},${y})}${text}`;
+}
+
+test('parseSubtitleCues joins caption rows that wrap one sentence across two events', () => {
+  const content = [
+    ...captionRowsHeader,
+    captionRow('0:00:19.08', '0:00:22.66', 172, 437, '⸨ぶっちゃけ'),
+    captionRow(
+      '0:00:19.08',
+      '0:00:22.66',
+      172,
+      497,
+      '早く{\\fscx50}　{\\fscx100}この勉強生活　終えたいし⸩',
+    ),
+    // No bracket at all: the upper row simply has not reached sentence punctuation.
+    captionRow('0:02:42.33', '0:02:44.43', 232, 437, '（東）≪好きだと'),
+    captionRow('0:02:42.33', '0:02:44.43', 232, 497, '自覚してしまったものの➡'),
+    // Rows are centred independently, so a wrap can change x between rows.
+    captionRow('0:00:42.21', '0:00:45.21', 252, 407, '≪ちょっとしたことで'),
+    captionRow('0:00:42.21', '0:00:45.21', 292, 497, '勝手に落ち込んだり➡'),
+    // A quote closed with 」 inside a still-open ≪…≫ span is not the end of the line.
+    captionRow('0:19:02.84', '0:19:05.00', 212, 437, '≪「つきあえる自信がない」'),
+    captionRow('0:19:02.84', '0:19:05.00', 452, 497, 'じゃない≫'),
+    // An in-sentence 「 quote on the lower row is not a new turn.
+    captionRow('0:18:35.55', '0:18:38.00', 232, 437, '今　「好きだ」と'),
+    captionRow('0:18:35.55', '0:18:38.00', 192, 497, '「心地いい」と感じてるのも➡'),
+  ].join('\n');
+
+  const cues = parseSubtitleCues(content, 'polar-opposites-s02e09.ass');
+
+  assert.deepEqual(
+    cues.map((cue) => cue.text),
+    [
+      '⸨ぶっちゃけ\n早く　この勉強生活　終えたいし⸩',
+      '≪ちょっとしたことで\n勝手に落ち込んだり➡',
+      '（東）≪好きだと\n自覚してしまったものの➡',
+      '今　「好きだ」と\n「心地いい」と感じてるのも➡',
+      '≪「つきあえる自信がない」\nじゃない≫',
+    ],
+  );
+  assert.ok(cues.every((cue) => cue.assLayout?.kind === 'positioned'));
+});
+
+test('parseSubtitleCues keeps simultaneous caption rows from two speakers separate', () => {
+  const content = [
+    ...captionRowsHeader,
+    // Both unlabeled: the upper row finished its sentence.
+    captionRow('0:03:56.10', '0:04:00.04', 172, 437, 'なあ　車両　変えね？'),
+    captionRow('0:03:56.10', '0:04:00.04', 632, 497, 'えっ？➡'),
+    // Lower row opens a labeled turn.
+    captionRow('0:03:38.48', '0:03:42.05', 592, 437, 'おはよう！'),
+    captionRow('0:03:38.48', '0:03:42.05', 272, 497, '（平）あっ　声　でかっ。'),
+    // A closed monologue span above a sound effect.
+    captionRow('0:08:16.83', '0:08:19.50', 372, 437, '≪落ち着け　落ち着け≫'),
+    captionRow('0:08:16.83', '0:08:19.50', 272, 497, 'ドクン　ドクン　ドクン…'),
+    // Two labeled speakers.
+    captionRow('0:09:27.90', '0:09:31.07', 312, 437, '（平）ぐぅ…。'),
+    captionRow('0:09:27.90', '0:09:31.07', 352, 497, '（東）≪ちくしょう～！≫'),
+    // A bare label never swallows a differently labeled row.
+    captionRow('0:11:43.24', '0:11:45.00', 212, 437, '（長谷川）'),
+    captionRow('0:11:43.24', '0:11:45.00', 412, 497, '（早乙女）ん？'),
+    // A short sentence-final 。 closes the upper row like any other.
+    captionRow('0:12:31.55', '0:12:33.55', 172, 437, '⚞（東）平。'),
+    captionRow('0:12:31.55', '0:12:33.55', 532, 497, 'あっ。'),
+  ].join('\n');
+
+  const cues = parseSubtitleCues(content, 'polar-opposites-s02e09.ass');
+
+  assert.deepEqual(
+    cues.map((cue) => cue.text),
+    [
+      'おはよう！',
+      '（平）あっ　声　でかっ。',
+      'なあ　車両　変えね？',
+      'えっ？➡',
+      '≪落ち着け　落ち着け≫',
+      'ドクン　ドクン　ドクン…',
+      '（平）ぐぅ…。',
+      '（東）≪ちくしょう～！≫',
+      '（長谷川）',
+      '（早乙女）ん？',
+      '⚞（東）平。',
+      'あっ。',
+    ],
+  );
+});
+
+test('parseSubtitleCues keeps caption rows apart across styles, bands, and timing', () => {
+  const content = [
+    ...captionRowsHeader,
+    // Same wording as a wrap, but the rows sit in different vertical bands.
+    captionRow('0:01:00.00', '0:01:02.00', 172, 77, '≪ちょっとしたことで'),
+    captionRow('0:01:00.00', '0:01:02.00', 172, 497, '勝手に落ち込んだり➡'),
+    // Same band, but a sign style beside dialogue.
+    'Dialogue: 0,0:01:05.00,0:01:07.00,Sign,,0,0,0,,{\\pos(172,437)}ちょっとしたことで',
+    captionRow('0:01:05.00', '0:01:07.00', 172, 497, '勝手に落ち込んだり➡'),
+    // Same rows, but the lower one ends later.
+    captionRow('0:01:10.00', '0:01:12.00', 172, 437, '≪ちょっとしたことで'),
+    captionRow('0:01:10.00', '0:01:13.00', 172, 497, '勝手に落ち込んだり➡'),
+    // Style-aligned rows without \pos are never caption rows.
+    'Dialogue: 0,0:01:15.00,0:01:17.00,Default,,0,0,0,,{\\an8}≪ちょっとしたことで',
+    'Dialogue: 0,0:01:15.00,0:01:17.00,Default,,0,0,0,,{\\an2}勝手に落ち込んだり➡',
+    // Same height: the events sit side by side, not one above the other.
+    captionRow('0:01:20.00', '0:01:22.00', 172, 497, '≪ちょっとしたことで'),
+    captionRow('0:01:20.00', '0:01:22.00', 612, 497, '勝手に落ち込んだり➡'),
+    // Same bottom band, but further apart than two text rows.
+    captionRow('0:01:25.00', '0:01:27.00', 172, 367, '≪ちょっとしたことで'),
+    captionRow('0:01:25.00', '0:01:27.00', 172, 497, '勝手に落ち込んだり➡'),
+  ].join('\n');
+
+  const cues = parseSubtitleCues(content, 'test.ass');
+
+  assert.equal(cues.length, 12);
+  assert.ok(cues.every((cue) => !cue.text.includes('\n')));
+});
+
+test('parseSubtitleCues leaves typeset rows alone in scripts that are not broadcast captions', () => {
+  // Fansub typesetting stacks positioned rows for signs, chat bubbles, and headlines. Such
+  // text carries no caption punctuation, so without the script-level gate every stacked
+  // pair here would read as an unfinished sentence and merge.
+  const content = [
+    ...captionRowsHeader,
+    captionRow('0:00:10.00', '0:00:14.00', 640, 200, 'Shocking Statement Leaves'),
+    captionRow('0:00:10.00', '0:00:14.00', 640, 260, 'Listeners Speechless!'),
+    captionRow('0:01:00.00', '0:01:04.00', 400, 300, 'shes here AGAIN'),
+    captionRow('0:01:00.00', '0:01:04.00', 400, 360, 'make sakiko-chan go home'),
+    // Japanese typesetting in the same script is held back by the same gate.
+    captionRow('0:02:00.00', '0:02:04.00', 300, 400, '定休日'),
+    captionRow('0:02:00.00', '0:02:04.00', 300, 460, '毎週水曜日'),
+  ].join('\n');
+
+  const cues = parseSubtitleCues(content, 'test.ass');
+
+  assert.deepEqual(
+    cues.map((cue) => cue.text),
+    [
+      'Shocking Statement Leaves',
+      'Listeners Speechless!',
+      'shes here AGAIN',
+      'make sakiko-chan go home',
+      '定休日',
+      '毎週水曜日',
+    ],
+  );
+});
+
+test('parseSubtitleCues never joins caption rows that carry no Japanese', () => {
+  // Even inside a caption script, romaji or English rows are not the wrapped Japanese
+  // sentences this pass targets.
+  const content = [
+    ...captionRowsHeader,
+    captionRow('0:00:10.00', '0:00:13.00', 172, 437, '（東）≪好きだと'),
+    captionRow('0:00:10.00', '0:00:13.00', 172, 497, '自覚してしまったものの➡'),
+    captionRow('0:00:20.00', '0:00:23.00', 172, 437, '（平）ん？'),
+    captionRow('0:00:30.00', '0:00:34.00', 640, 437, 'NOW LOADING'),
+    captionRow('0:00:30.00', '0:00:34.00', 640, 497, 'please wait'),
+  ].join('\n');
+
+  const cues = parseSubtitleCues(content, 'test.ass');
+
+  assert.deepEqual(
+    cues.map((cue) => cue.text),
+    ['（東）≪好きだと\n自覚してしまったものの➡', '（平）ん？', 'NOW LOADING', 'please wait'],
+  );
 });
 
 test('parseSubtitleCues scales furigana geometry by PlayResY', () => {

src/core/services/subtitle-cue-parser.ts

@@ -2342,6 +2342,185 @@ function removeAssFuriganaEvents(
   };
 }
 
+// Broadcast-caption converters give every visual row of one utterance its own positioned
+// event, so a sentence that wraps arrives as two simultaneous cues with the same timing,
+// style, and vertical band. Captions punctuate every finished utterance, and each turn
+// opens with a speaker label or a ≪…≫ / ⸨…⸩ span, which is what tells a wrapped sentence
+// apart from two speakers sharing the screen.
+const CAPTION_SPEAKER_LABEL_ONLY_PATTERN = /^（[^（）]*）$/u;
+const CAPTION_SPEAKER_LABEL_PATTERN = /^（/u;
+const CAPTION_TURN_OPENER_PATTERN = /^[≪⸨（]/u;
+const CAPTION_TERMINAL_PATTERN = /[。？！?!…‥～〜➡⁉⁈≫⸩）」』]$/u;
+const CAPTION_SPANS: ReadonlyArray<readonly [open: string, close: string]> = [
+  ['≪', '≫'],
+  ['⸨', '⸩'],
+];
+// Rows of one utterance sit one text row apart (about 60 units in the 540-line space the
+// furigana geometry is tuned for), or two when a ruby row lies between them. Rows at the
+// same height sit side by side, and rows further apart are separate placements.
+const MAX_CAPTION_ROW_GAP = 120;
+// Only a broadcast-caption script gets rows joined. Typesetters position rows for signs,
+// chat bubbles, and lyric stacks too, and there the continuation rule below has no
+// convention to read: sign text rarely carries sentence punctuation, so unrelated rows
+// would run together. A caption script announces itself by labelling speakers （名） and
+// bracketing off-screen speech in ≪…≫ / ⸨…⸩; typeset scripts use those in a handful of
+// lines at most. Measured over local tracks, caption scripts sit near 25% and every typeset
+// script below 1%, so the threshold has room on both sides. It is deliberately strict: a
+// caption script wrongly held back just keeps one sentence on two rows, while a typeset
+// script wrongly let through concatenates unrelated signs.
+const MIN_CAPTION_EVIDENCE_EVENTS = 2;
+const MIN_CAPTION_EVIDENCE_RATIO = 0.05;
+const CAPTION_EVIDENCE_PATTERN = /^（[^（）]{1,14}）|[≪⸨]/u;
+// Rows that carry no Japanese are not the broadcast captions this pass targets.
+const JAPANESE_SCRIPT_PATTERN = /[\p{Script=Hiragana}\p{Script=Katakana}\p{Script=Han}]/u;
+
+function hasBroadcastCaptionConventions(cues: readonly AnnotatedSubtitleCue[]): boolean {
+  let evidence = 0;
+  let published = 0;
+  for (const cue of cues) {
+    if (!cue.text.trim()) continue;
+    published += 1;
+    if (CAPTION_EVIDENCE_PATTERN.test(cue.text)) evidence += 1;
+  }
+  return (
+    evidence >= MIN_CAPTION_EVIDENCE_EVENTS && evidence >= published * MIN_CAPTION_EVIDENCE_RATIO
+  );
+}
+
+function captionSpanDepth(text: string, [open, close]: readonly [string, string]): number {
+  let depth = 0;
+  for (const char of text) {
+    if (char === open) depth += 1;
+    else if (char === close) depth -= 1;
+  }
+  return depth;
+}
+
+/**
+ * Whether `lower` continues the utterance `upper` started, both being simultaneous
+ * caption rows. A bare speaker label labels the row beneath it. Otherwise the upper row
+ * must not have finished: it ends without terminal punctuation, or a ≪…≫ / ⸨…⸩ span it
+ * opened is still open (closing 」 inside such a span is not an ending). A lower row that
+ * opens its own turn is always a different line.
+ */
+function isCaptionRowContinuation(upper: string, lower: string): boolean {
+  if (CAPTION_SPEAKER_LABEL_ONLY_PATTERN.test(upper)) {
+    return !CAPTION_SPEAKER_LABEL_PATTERN.test(lower);
+  }
+  if (CAPTION_TURN_OPENER_PATTERN.test(lower)) {
+    return false;
+  }
+  const spanContinues = CAPTION_SPANS.some(
+    (span) => captionSpanDepth(upper, span) > 0 || captionSpanDepth(lower, span) < 0,
+  );
+  return spanContinues || !CAPTION_TERMINAL_PATTERN.test(upper);
+}
+
+// A half-height row is ruby or a whispered aside, not a row of the utterance.
+function isCaptionRowCandidate(cue: AnnotatedSubtitleCue): boolean {
+  const scaleY = staticAssScalePercent(cue, 'fscy');
+  return (
+    cue.source === undefined &&
+    cue.assLayout?.kind === 'positioned' &&
+    cue.effect.trim() === '' &&
+    !cue.text.includes('\n') &&
+    !hasAssTemporalOverride(cue.overrides) &&
+    (scaleY === null || scaleY > MAX_ASS_FURIGANA_SCALE_PERCENT) &&
+    JAPANESE_SCRIPT_PATTERN.test(cue.text)
+  );
+}
+
+function captionRowGroupKey(cue: AnnotatedSubtitleCue): string {
+  return [
+    cue.startTime,
+    cue.endTime,
+    cue.style,
+    cue.layer,
+    cue.name,
+    cue.assLayout?.verticalBand ?? '',
+  ].join('\0');
+}
+
+function mergeCaptionRows(rows: readonly AnnotatedSubtitleCue[]): AnnotatedSubtitleCue {
+  const [first] = rows;
+  if (!first) throw new Error('mergeCaptionRows requires at least one row');
+  const overrides = rows.flatMap((row) => row.overrides);
+  const assFurigana = [...new Set(rows.flatMap((row) => row.assFurigana ?? []))];
+  return {
+    ...first,
+    text: rows.map((row) => row.text).join('\n'),
+    rawText: rows.map((row) => row.rawText).join('\\N'),
+    overrides,
+    overrideSignature: assOverrideSignature(overrides),
+    ...(assFurigana.length === 0 ? {} : { assFurigana }),
+  };
+}
+
+/**
+ * Join simultaneous caption rows that spell one utterance into a single cue, so the
+ * overlay can wrap or flatten it like an authored `\N` line and the sidebar and mining
+ * paths see the whole sentence. Rows stack top to bottom; each row joins the cue above it
+ * only while `isCaptionRowContinuation` holds, so a second speaker starts a new cue. The
+ * whole pass is skipped unless the script reads as broadcast captions.
+ */
+function mergeAssCaptionRows(
+  cues: AnnotatedSubtitleCue[],
+  playResY: number | null,
+): AnnotatedSubtitleCue[] {
+  if (!hasBroadcastCaptionConventions(cues)) return cues;
+
+  const groups = new Map<string, AnnotatedSubtitleCue[]>();
+  for (const cue of cues) {
+    if (!isCaptionRowCandidate(cue)) continue;
+    const key = captionRowGroupKey(cue);
+    const group = groups.get(key);
+    if (group) group.push(cue);
+    else groups.set(key, [cue]);
+  }
+
+  const maxRowGap = MAX_CAPTION_ROW_GAP * assFuriganaGeometryScale(playResY);
+  const rowY = (cue: AnnotatedSubtitleCue): number =>
+    cue.assLayout?.kind === 'positioned' ? cue.assLayout.y : 0;
+  const replacements = new Map<AnnotatedSubtitleCue, AnnotatedSubtitleCue>();
+  const removed = new Set<AnnotatedSubtitleCue>();
+  for (const group of groups.values()) {
+    if (group.length < 2) continue;
+    const rows = [...group].sort((a, b) => rowY(a) - rowY(b) || a.order - b.order);
+    let run: AnnotatedSubtitleCue[] = [];
+    const flush = (): void => {
+      if (run.length < 2) return;
+      const anchor = run.reduce((lowest, row) => (row.order < lowest.order ? row : lowest));
+      replacements.set(anchor, mergeCaptionRows(run));
+      for (const row of run) {
+        if (row !== anchor) removed.add(row);
+      }
+    };
+    for (const row of rows) {
+      const previous = run.at(-1);
+      const gap = previous ? rowY(row) - rowY(previous) : 0;
+      if (
+        previous &&
+        gap > 0 &&
+        gap <= maxRowGap &&
+        previous.text !== row.text &&
+        isCaptionRowContinuation(previous.text, row.text)
+      ) {
+        run.push(row);
+        continue;
+      }
+      flush();
+      run = [row];
+    }
+    flush();
+  }
+
+  if (replacements.size === 0) return cues;
+  return cues.flatMap((cue) => {
+    if (removed.has(cue)) return [];
+    return [replacements.get(cue) ?? cue];
+  });
+}
+
 function parseAnnotatedAssEvents(content: string, placement: AssPlacementContext): ParsedAssEvents {
   const cues: AnnotatedSubtitleCue[] = [];
   const comments: AnnotatedSubtitleCue[] = [];
@@ -2476,7 +2655,10 @@ function parseAnnotatedAssCues(content: string): AnnotatedSubtitleCue[] {
     removeAssFontTextureEvents(parseAnnotatedAssEvents(content, placement)),
     placement.playResY,
   );
-  return recoverFragmentOnlyAssLines(recoverCanonicalAssEvents(events));
+  return mergeAssCaptionRows(
+    recoverFragmentOnlyAssLines(recoverCanonicalAssEvents(events)),
+    placement.playResY,
+  );
 }
 
 export function parseAssCues(content: string): SubtitleCue[] {

src/main/runtime/primary-subtitle-text.test.ts

@@ -674,3 +674,31 @@ test('resolvePrimarySubtitleText keeps source order when no cue declares a place
     'First line\n\nSecond line',
   );
 });
+
+test('resolvePrimarySubtitleText publishes a wrapped caption sentence as one cue', () => {
+  // mpv still reports the two source rows (plus the ruby row) as separate live lines, so
+  // the merged cue must explain all of them and come back as a single-break line the
+  // display layer may flatten, not as a two-cue boundary.
+  const ass = [
+    '[Script Info]',
+    'PlayResY: 540',
+    '',
+    '[Events]',
+    'Format: Layer, Start, End, Style, Name, MarginL, MarginR, MarginV, Effect, Text',
+    'Dialogue: 0,0:02:42.33,0:02:44.43,Default,,0,0,0,,{\\pos(232,437)\\fscx50}（{\\fscx100}東{\\fscx50}）{\\fscx100}≪好きだと',
+    'Dialogue: 0,0:02:42.33,0:02:44.43,Default,,0,0,0,,{\\pos(292,443)\\fscx50\\fscy50}じかく',
+    'Dialogue: 0,0:02:42.33,0:02:44.43,Default,,0,0,0,,{\\pos(232,497)}自覚してしまったものの➡',
+    // A second labeled turn, so the script reads as broadcast captions.
+    'Dialogue: 0,0:02:44.43,0:02:47.37,Default,,0,0,0,,{\\pos(212,497)\\fscx50}（{\\fscx100}平{\\fscx50}）{\\fscx100}どうした？',
+  ].join('\n');
+  const cues = parseSubtitleCues(ass, 'polar-opposites-s02e09.ass');
+
+  assert.equal(
+    resolvePrimarySubtitleText({
+      liveText: '（東）≪好きだと\nじかく\n自覚してしまったものの➡',
+      currentTimeSec: 163,
+      cues,
+    }),
+    '（東）≪好きだと\n自覚してしまったものの➡',
+  );
+});