sudacode/SubMiner
1
0
Fork 0
mirror of https://github.com/ksyasuda/SubMiner.git synced 2026-08-26 00:15:27 -07:00
Code Issues Packages Projects Releases Wiki Activity

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

Browse Source 
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.
This commit is contained in:
sudacode
2026-08-26 00:07:42 -07:00
parent c2c25c0da6
commit f8ca8681dc
3 changed files with 78 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
changes/youtube-auto-caption-duration.md
+4
View File
@@ -0,0 +1,4 @@
type: fixed
area: youtube
- YouTube auto-generated captions no longer disappear mid-speech: long caption rows carry a placeholder 3-second duration in YouTube's TimedText data, so converted cues now stay on screen until the next caption event, matching YouTube's own display timing.
src/core/services/youtube/timedtext.test.ts
+30
View File
@@ -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', () => { test('normalizeYoutubeAutoVtt strips cumulative rolling-caption prefixes', () => {
const result = normalizeYoutubeAutoVtt( const result = normalizeYoutubeAutoVtt(
[ [
src/core/services/youtube/timedtext.ts
+44 -7
View File
@@ -4,6 +4,16 @@ interface YoutubeTimedTextRow {
text: string; text: string;
} }
interface YoutubeTimedTextDocument {
rows: YoutubeTimedTextRow[];
// Start times of every <p> event, including empty window-append fillers.
// In the rolling auto-caption format YouTube displays each caption until the
// next window event and the row's own duration is often a 3000ms placeholder,
// so these timestamps are the only reliable source for cue end times.
eventStartsMs: number[];
hasRollingWindowEvents: boolean;
}
const YOUTUBE_TIMEDTEXT_EXTENSIONS = new Set(['srv1', 'srv2', 'srv3', 'ytsrv3']); const YOUTUBE_TIMEDTEXT_EXTENSIONS = new Set(['srv1', 'srv2', 'srv3', 'ytsrv3']);
function decodeNumericEntity(match: string, codePoint: number): string { function decodeNumericEntity(match: string, codePoint: number): string {
@@ -39,14 +49,24 @@ function parseAttributeMap(raw: string): Map<string, string> {
return attrs; return attrs;
} }
function extractYoutubeTimedTextRows(xml: string): YoutubeTimedTextRow[] { function extractYoutubeTimedTextDocument(xml: string): YoutubeTimedTextDocument {
const rows: YoutubeTimedTextRow[] = []; const rows: YoutubeTimedTextRow[] = [];
const eventStartsMs: number[] = [];
let hasRollingWindowEvents = false;
for (const match of xml.matchAll(/<p\b([^>]*)>([\s\S]*?)<\/p>/g)) { for (const match of xml.matchAll(/<p\b([^>]*)>([\s\S]*?)<\/p>/g)) {
const attrs = parseAttributeMap(match[1] ?? ''); const attrs = parseAttributeMap(match[1] ?? '');
const startMs = Number(attrs.get('t')); const startMs = Number(attrs.get('t'));
if (!Number.isFinite(startMs)) {
continue;
}
eventStartsMs.push(startMs);
if (attrs.get('a') === '1') {
hasRollingWindowEvents = true;
}
const durationMs = Number(attrs.get('d')); const durationMs = Number(attrs.get('d'));
if (!Number.isFinite(startMs) || !Number.isFinite(durationMs)) { if (!Number.isFinite(durationMs)) {
continue; continue;
} }
@@ -59,7 +79,17 @@ function extractYoutubeTimedTextRows(xml: string): YoutubeTimedTextRow[] {
rows.push({ startMs, durationMs, text }); rows.push({ startMs, durationMs, text });
} }
return rows; eventStartsMs.sort((a, b) => a - b);
return { rows, eventStartsMs, hasRollingWindowEvents };
}
function findNextEventStartMs(eventStartsMs: number[], afterMs: number): number | undefined {
for (const startMs of eventStartsMs) {
if (startMs > afterMs) {
return startMs;
}
}
return undefined;
} }
function formatVttTimestamp(ms: number): string { function formatVttTimestamp(ms: number): string {
@@ -79,7 +109,7 @@ export function isYoutubeTimedTextExtension(value: string | undefined): boolean
} }
export function convertYoutubeTimedTextToVtt(xml: string): string { export function convertYoutubeTimedTextToVtt(xml: string): string {
const rows = extractYoutubeTimedTextRows(xml); const { rows, eventStartsMs, hasRollingWindowEvents } = extractYoutubeTimedTextDocument(xml);
if (rows.length === 0) { if (rows.length === 0) {
return 'WEBVTT\n'; return 'WEBVTT\n';
} }
@@ -90,10 +120,17 @@ export function convertYoutubeTimedTextToVtt(xml: string): string {
const row = rows[index]!; const row = rows[index]!;
const nextRow = rows[index + 1]; const nextRow = rows[index + 1];
const unclampedEnd = row.startMs + row.durationMs; const unclampedEnd = row.startMs + row.durationMs;
// Rolling auto captions display until the next window event; the row's own
// duration is frequently a 3000ms placeholder that cuts long lines short.
const nextEventStart = hasRollingWindowEvents
? findNextEventStartMs(eventStartsMs, row.startMs)
: undefined;
const clampedEnd = const clampedEnd =
nextRow && unclampedEnd > nextRow.startMs nextEventStart !== undefined
? Math.max(row.startMs, nextRow.startMs - 1) ? nextEventStart
: unclampedEnd; : nextRow && unclampedEnd > nextRow.startMs
? Math.max(row.startMs, nextRow.startMs - 1)
: unclampedEnd;
if (clampedEnd <= row.startMs) { if (clampedEnd <= row.startMs) {
continue; continue;
} }