Files
SubMiner/src/main/runtime/primary-subtitle-text.ts
T
sudacode c01bcd9d0f fix(subtitles): collapse layered ASS lyrics across startup and seeks
- Reconstruct fragment-only karaoke while preserving canonical animated signs
- Reprocess live subtitles after initial playback and seek updates
- Add a development launcher for playable media files
2026-08-19 22:55:10 -07:00

223 lines
8.0 KiB
TypeScript

import type { SubtitleCue } from '../../types';
// 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
// entrance/exit frames actually run past the authored timing.
const LIVE_CUE_EDGE_TOLERANCE_SECONDS = 1;
export interface ResolvedPrimarySubtitle {
text: string;
startTime: number;
endTime: number;
/** The parsed cues behind `text`, for consumers that record lines individually. */
cues: SubtitleCue[];
}
function animationSpan(cue: SubtitleCue): { start: number; end: number } {
return {
start: cue.animationStartTime ?? cue.startTime,
end: cue.animationEndTime ?? cue.endTime,
};
}
function nearbyCanonicalCues(
cues: readonly SubtitleCue[] | null | undefined,
currentTimeSec: number,
): SubtitleCue[] {
return (cues ?? []).filter((cue) => {
if (cue.source !== 'canonical-ass' && cue.source !== 'reconstructed-ass') {
return false;
}
const span = animationSpan(cue);
return (
span.end >= currentTimeSec - LIVE_CUE_EDGE_TOLERANCE_SECONDS &&
span.start <= currentTimeSec + LIVE_CUE_EDGE_TOLERANCE_SECONDS
);
});
}
function compactWhitespace(text: string): string {
return text.replace(/\s+/gu, '');
}
// ASS layers can encode the same visible spacing with ordinary, hard, or
// ideographic spaces. Matching and emission must use the same identity or each
// layer reappears as a copy.
function uniqueCueTexts(cues: readonly SubtitleCue[]): string[] {
const texts: string[] = [];
const seen = new Set<string>();
for (const cue of cues) {
const compactText = compactWhitespace(cue.text);
if (seen.has(compactText)) continue;
seen.add(compactText);
texts.push(cue.text);
}
return texts;
}
function compactLineSegments(text: string): string[] {
return text.split('\n').map(compactWhitespace).filter(Boolean);
}
/**
* Parsed cues have already collapsed exact ASS layers and animation runs. Trust that
* cleaner view only when every live mpv line is accounted for by an active parsed cue.
* This keeps unrelated concurrent dialogue on the live fallback while removing style
* stacks where mpv repeats one full lyric for fill, border, blur, and shadow layers.
*/
function resolveActiveParsedPrimarySubtitle(options: {
liveText: string;
currentTimeSec: number;
cues: readonly SubtitleCue[] | null | undefined;
}): ResolvedPrimarySubtitle | null {
if (!Number.isFinite(options.currentTimeSec)) {
return null;
}
const liveSegments = compactLineSegments(options.liveText);
if (liveSegments.length === 0) {
return null;
}
const liveSegmentSet = new Set(liveSegments);
const selected = (options.cues ?? []).filter((cue) => {
if (
cue.startTime > options.currentTimeSec + LIVE_CUE_EDGE_TOLERANCE_SECONDS ||
cue.endTime <= options.currentTimeSec - LIVE_CUE_EDGE_TOLERANCE_SECONDS
) {
return false;
}
const cueSegments = compactLineSegments(cue.text);
return cueSegments.length > 0 && cueSegments.every((segment) => liveSegmentSet.has(segment));
});
if (selected.length === 0) {
return null;
}
const parsedSegmentSet = new Set(selected.flatMap((cue) => compactLineSegments(cue.text)));
if (!liveSegments.every((segment) => parsedSegmentSet.has(segment))) {
return null;
}
const texts = uniqueCueTexts(selected);
return {
text: texts.join('\n'),
startTime: Math.min(...selected.map((cue) => cue.startTime)),
endTime: Math.max(...selected.map((cue) => cue.endTime)),
cues: selected,
};
}
/**
* mpv's `sub-text` renders each simultaneously active ASS event on its own line, so
* while a generated animation plays every live line is a contiguous piece of the
* authored text. A line that is not -- concurrent dialogue during an insert song, or a
* fresh line starting just after the animation ended -- proves the live text is not this
* animation, and substituting the canonical line would swallow real dialogue.
*/
function liveTextIsFromCues(liveText: string, cues: readonly SubtitleCue[]): boolean {
const compactCues = cues.map((cue) => compactWhitespace(cue.text));
const segments = liveText.split('\n').map(compactWhitespace).filter(Boolean);
return (
segments.length > 0 &&
segments.every((segment) => compactCues.some((cueText) => cueText.includes(segment)))
);
}
export function resolveCanonicalPrimarySubtitle(options: {
liveText: string;
currentTimeSec: number;
cues: readonly SubtitleCue[] | null | undefined;
}): ResolvedPrimarySubtitle | null {
if (!Number.isFinite(options.currentTimeSec)) {
return null;
}
// Consecutive karaoke lines overlap: one line's exit frames are still on screen while
// the next line's entrance frames appear. The fragment check therefore runs against
// every canonical cue whose animation envelope reaches the current time, while only
// the active (or single nearest) cue supplies the displayed text.
const nearby = nearbyCanonicalCues(options.cues, options.currentTimeSec);
const active = nearby.filter(
(cue) => cue.startTime <= options.currentTimeSec && cue.endTime > options.currentTimeSec,
);
const liveSegments = options.liveText.split('\n').map(compactWhitespace).filter(Boolean);
const selected =
active.length > 0
? active
: nearby
// Between authored spans, proximity alone can pick the wrong neighbor: the
// next line can sit closer while only the previous line's exit fragments are
// on screen. Only cues that explain at least one live line may be selected.
.filter((cue) => {
const cueText = compactWhitespace(cue.text);
return liveSegments.some((segment) => cueText.includes(segment));
})
.map((cue) => {
const span = animationSpan(cue);
const distance =
options.currentTimeSec < span.start
? span.start - options.currentTimeSec
: Math.max(0, options.currentTimeSec - span.end);
return { cue, distance };
})
.sort((a, b) => a.distance - b.distance || a.cue.startTime - b.cue.startTime)
.slice(0, 1)
.map(({ cue }) => cue);
if (selected.length === 0 || !liveTextIsFromCues(options.liveText, nearby)) {
return null;
}
const texts = uniqueCueTexts(selected);
return {
text: texts.join('\n'),
startTime: Math.min(...selected.map((cue) => cue.startTime)),
endTime: Math.max(...selected.map((cue) => cue.endTime)),
cues: selected,
};
}
/**
* Live text with generated-animation fragment lines removed. Recording paths use this
* when full canonical substitution declined -- concurrent dialogue during an insert
* song: the dialogue is worth recording, the glyph fragments beside it are not. Returns
* the input unchanged when no canonical cue is near or nothing non-fragment remains.
*/
export function stripCanonicalFragmentLines(options: {
liveText: string;
currentTimeSec: number;
cues: readonly SubtitleCue[] | null | undefined;
}): string {
if (!Number.isFinite(options.currentTimeSec)) {
return options.liveText;
}
const nearby = nearbyCanonicalCues(options.cues, options.currentTimeSec);
if (nearby.length === 0) {
return 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));
});
return kept.length > 0 ? kept.join('\n') : options.liveText;
}
export function resolvePrimarySubtitleText(options: {
liveText: string;
currentTimeSec: number;
cues: readonly SubtitleCue[] | null | undefined;
}): string {
if (!options.liveText.trim()) {
return options.liveText;
}
return (
resolveCanonicalPrimarySubtitle({
liveText: options.liveText,
currentTimeSec: options.currentTimeSec,
cues: options.cues,
})?.text ??
resolveActiveParsedPrimarySubtitle(options)?.text ??
options.liveText
);
}