fix(subtitles): suppress per-glyph typesetting walls in live subtitle text

When embedded-track extraction is skipped (network-mounted media), live mpv
text during per-glyph typeset karaoke is a wall of simultaneous one-glyph
lines plus the syllable being typed. No parsed cues exist to substitute, so
the wall reached both overlays and recording verbatim. Detect bursts of many
single-glyph lines in the live fallback paths and drop them with their short
syllable companions, keeping concurrent dialogue lines.

changes/dedupe-embedded-subtitle-extraction.md

@@ -2,3 +2,4 @@ type: fixed
 area: subtitles
 
 - Prevented embedded subtitle parsing from starving network playback: mounted SMB/NFS media now uses deduplicated mpv live text, while duplicate extraction requests for local media share one ffmpeg process.
+- Live subtitle text from per-glyph typeset karaoke (network-mounted media without parsed cues) no longer shows a wall of scattered letters in the overlays; the glyph wall and its typed-syllable fragments are suppressed while concurrent dialogue lines remain.

docs/architecture/subtitle-overlay-priming.md

@@ -100,6 +100,11 @@ coming and prefetching would otherwise idle for the rest of the cue.
 - `secondary-sub-text` remains the immediate fallback, so unreadable subtitle sources, remote URLs,
   and files on network mounts still appear without waiting for file resolution. Embedded-track
   extraction is skipped for those sources to avoid competing with playback for network bandwidth.
+- The live fallback also suppresses per-glyph typesetting walls: when many simultaneous
+  one-glyph lines are present (generated karaoke lettering flattened into live text), those
+  lines and their short syllable companions are dropped while concurrent dialogue lines stay.
+  This covers network-mounted media, where embedded-track extraction is skipped and no parsed
+  cues exist to substitute.
 - Parsed secondary text and the live fallback remove exact repeated lines at any length. A
   flattened-line identity also removes long dialogue/sign repetitions that differ only in
   whitespace or terminal punctuation, while distinct simultaneous short lines remain separate.

src/core/services/ass-text.test.ts

@@ -10,6 +10,7 @@ import {
   isAssTemporalCommand,
   normalizePlainSubtitleText,
   parseAssEffectField,
+  removeLiveGlyphFragmentLines,
   removeAssControlDebrisLines,
 } from './ass-text';
 
@@ -202,3 +203,18 @@ test('isAnimatedAssEffectKind covers the stock animated effects only', () => {
   assert.equal(isAnimatedAssEffectKind('other'), false);
   assert.equal(isAnimatedAssEffectKind('none'), false);
 });
+
+test('removeLiveGlyphFragmentLines drops a per-glyph typesetting wall and its syllable', () => {
+  const wall = [...'wansdumretoikhI'].join('\n');
+  assert.equal(removeLiveGlyphFragmentLines(`${wall}\ntai`), '');
+});
+
+test('removeLiveGlyphFragmentLines keeps concurrent dialogue beside a glyph wall', () => {
+  const wall = [...'wansdumretoikhI'].join('\n');
+  assert.equal(removeLiveGlyphFragmentLines(`${wall}\nそれよりも　ノート…`), 'それよりも　ノート…');
+});
+
+test('removeLiveGlyphFragmentLines leaves ordinary short lines alone', () => {
+  const text = 'え\nはい。\nそうだな';
+  assert.equal(removeLiveGlyphFragmentLines(text), text);
+});

src/core/services/ass-text.ts

@@ -108,6 +108,24 @@ export function removeAssControlDebrisLines(text: string): string {
     .join('\n');
 }
 
+const MIN_GLYPH_BURST_LINES = 6;
+const MAX_GLYPH_BURST_COMPANION_GLYPHS = 3;
+
+/**
+ * Per-glyph karaoke typesetting flattened into live text becomes a wall of
+ * single-character lines plus the short syllable currently being typed. No authored
+ * subtitle stacks this many one-glyph lines at once, so when the wall is present drop
+ * it and its short companion fragments while keeping any concurrent dialogue line.
+ */
+export function removeLiveGlyphFragmentLines(text: string): string {
+  const lines = text.split('\n');
+  const singleGlyphLines = lines.filter((line) => [...line.trim()].length === 1).length;
+  if (singleGlyphLines < MIN_GLYPH_BURST_LINES) return text;
+  return lines
+    .filter((line) => [...line.trim()].length > MAX_GLYPH_BURST_COMPANION_GLYPHS)
+    .join('\n');
+}
+
 export interface NormalizePlainSubtitleTextOptions {
   /** Fold every line break into a single space. */
   collapseLineBreaks?: boolean;

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

@@ -477,3 +477,23 @@ test('resolveCanonicalPrimarySubtitle picks the cue its fragments spell, not the
     '今　手にある',
   );
 });
+
+test('resolvePrimarySubtitleText suppresses a live glyph wall when no cues are available', () => {
+  const wall = [...'wansdumretoikhI'].join('\n');
+  assert.equal(
+    resolvePrimarySubtitleText({ liveText: `${wall}\ntai`, currentTimeSec: 1355, cues: null }),
+    '',
+  );
+});
+
+test('stripCanonicalFragmentLines drops a live glyph wall with no nearby canonical cues', () => {
+  const wall = [...'wansdumretoikhI'].join('\n');
+  assert.equal(
+    stripCanonicalFragmentLines({
+      liveText: `${wall}\nそれよりも　ノート…`,
+      currentTimeSec: 1355,
+      cues: [],
+    }),
+    'それよりも　ノート…',
+  );
+});

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

@@ -1,5 +1,8 @@
 import type { SubtitleCue } from '../../types';
-import { removeAssControlDebrisLines } from '../../core/services/ass-text';
+import {
+  removeAssControlDebrisLines,
+  removeLiveGlyphFragmentLines,
+} from '../../core/services/ass-text';
 
 // Slack on top of each cue's recorded animation envelope, for time-pos observation
 // staleness and small user sub-delay offsets. The envelope itself covers how far
@@ -224,19 +227,20 @@ export function stripCanonicalFragmentLines(options: {
   cues: readonly SubtitleCue[] | null | undefined;
 }): string {
   if (!Number.isFinite(options.currentTimeSec)) {
-    return options.liveText;
+    return removeLiveGlyphFragmentLines(options.liveText);
   }
   const nearby = nearbyCanonicalCues(options.cues, options.currentTimeSec, true);
   if (nearby.length === 0) {
-    return options.liveText;
+    return removeLiveGlyphFragmentLines(options.liveText);
   }
   const compactCues = nearby.map((cue) => compactWhitespace(cue.text));
   const kept = options.liveText.split('\n').filter((line) => {
     const compact = compactWhitespace(line);
     return compact && !compactCues.some((cueText) => cueText.includes(compact));
   });
-  if (kept.length > 0) return kept.join('\n');
+  if (kept.length > 0) return removeLiveGlyphFragmentLines(kept.join('\n'));
-  return nearby.some((cue) => cue.assLayout?.kind === 'fragment-grid') ? '' : options.liveText;
+  if (nearby.some((cue) => cue.assLayout?.kind === 'fragment-grid')) return '';
+  return removeLiveGlyphFragmentLines(options.liveText);
 }
 
 export function resolvePrimarySubtitleText(options: {
@@ -257,6 +261,6 @@ export function resolvePrimarySubtitleText(options: {
       cues: options.cues,
     })?.text ??
     resolveActiveParsedPrimarySubtitle({ ...options, liveText })?.text ??
-    liveText
+    removeLiveGlyphFragmentLines(liveText)
   );
 }

src/main/runtime/secondary-subtitle-track.test.ts

@@ -581,3 +581,36 @@ test('secondary track controller ignores and cleans up a refresh invalidated by
   assert.equal(parseCalls, 0);
   assert.equal(cleanupCalls, 1);
 });
+
+test('secondary live fallback suppresses a per-glyph typesetting wall', async () => {
+  let currentText = '';
+  const controller = createSecondarySubtitleTrackController({
+    getMpvClient: () => ({
+      connected: true,
+      requestProperty: async (name) => {
+        if (name === 'secondary-sid') return 2;
+        if (name === 'track-list') return [{ type: 'sub', id: 2 }];
+        if (name === 'path') return '/mnt/nas/video.mkv';
+        if (name === 'secondary-sub-delay') return 0;
+        return null;
+      },
+    }),
+    getCurrentTimePos: () => 1355,
+    // Network-mounted media: embedded extraction is skipped, so no parsed cues exist.
+    resolveSubtitleSource: async () => null,
+    loadSubtitleSourceText: async () => '',
+    parseSubtitleCues,
+    setCurrentSecondaryText: (text) => {
+      currentText = text;
+    },
+    broadcastSecondaryText: () => {},
+  });
+
+  await controller.refresh();
+  const wall = [...'wansdumretoikhI'].join('\n');
+  controller.handleLiveText(`${wall}\ntai`);
+  assert.equal(currentText, '');
+
+  controller.handleLiveText(`${wall}\nそれよりも　ノート…`);
+  assert.equal(currentText, 'それよりも　ノート…');
+});

src/main/runtime/secondary-subtitle-track.ts

@@ -1,6 +1,9 @@
 import type { SubtitleCue } from '../../types/subtitle';
 import { flattenedSecondarySubtitleLineIdentity } from '../../core/services/secondary-subtitle-line-identity';
-import { removeAssControlDebrisLines } from '../../core/services/ass-text';
+import {
+  removeAssControlDebrisLines,
+  removeLiveGlyphFragmentLines,
+} from '../../core/services/ass-text';
 
 type SecondarySubtitleMpvClient = {
   connected?: boolean;
@@ -320,7 +323,9 @@ export function createSecondarySubtitleTrackController(deps: {
     refresh,
     scheduleRefresh,
     handleLiveText(text: string): void {
-      lastLiveText = activeSourceUsesAssSyntax ? removeAssControlDebrisLines(text) : text;
+      lastLiveText = removeLiveGlyphFragmentLines(
+        activeSourceUsesAssSyntax ? removeAssControlDebrisLines(text) : text,
+      );
       publish(resolveAtTime(deps.getCurrentTimePos()));
     },
     handleTimePos(timeSeconds: number): void {