mirror of
https://github.com/ksyasuda/SubMiner.git
synced 2026-09-12 05:16:19 -07:00
feat(anki): add Senren scene-switching field grouping
- Support auto, manual, and disabled Senren duplicate-card merges - Group sentence, furigana, audio, picture, and miscInfo fields
This commit is contained in:
@@ -86,6 +86,7 @@ const HOT_RELOAD_EXACT_OR_PREFIX_PATHS = [
|
||||
'ankiConnect.fields.miscInfo',
|
||||
'ankiConnect.isLapis.sentenceCardModel',
|
||||
'ankiConnect.isKiku.fieldGrouping',
|
||||
'ankiConnect.isSenren.fieldGrouping',
|
||||
'ankiConnect.lapisKiku.wordCardKind',
|
||||
] as const;
|
||||
|
||||
|
||||
@@ -131,12 +131,6 @@ export {
|
||||
resolvePlaybackPlan as resolveJellyfinPlaybackPlanRuntime,
|
||||
ticksToSeconds as jellyfinTicksToSecondsRuntime,
|
||||
} from './jellyfin';
|
||||
export { loadJellyfinSubtitleDelay, saveJellyfinSubtitleDelay } from './jellyfin-subtitle-delay';
|
||||
export {
|
||||
estimateSubtitleTimingOffset,
|
||||
type SubtitleTimingOffsetOptions,
|
||||
type SubtitleTimingOffsetResult,
|
||||
} from './subtitle-timing-offset';
|
||||
export { buildJellyfinTimelinePayload, JellyfinRemoteSessionService } from './jellyfin-remote';
|
||||
export {
|
||||
broadcastRuntimeOptionsChangedRuntime,
|
||||
|
||||
@@ -1,54 +0,0 @@
|
||||
import assert from 'node:assert/strict';
|
||||
import * as fs from 'node:fs';
|
||||
import * as os from 'node:os';
|
||||
import * as path from 'node:path';
|
||||
import test from 'node:test';
|
||||
import { loadJellyfinSubtitleDelay, saveJellyfinSubtitleDelay } from './jellyfin-subtitle-delay';
|
||||
|
||||
function statePath(name: string): string {
|
||||
return path.join(fs.mkdtempSync(path.join(os.tmpdir(), 'subminer-jellyfin-delay-')), name);
|
||||
}
|
||||
|
||||
test('jellyfin subtitle delay store saves and loads delay by item and stream', () => {
|
||||
const filePath = statePath('delays.json');
|
||||
|
||||
assert.equal(
|
||||
saveJellyfinSubtitleDelay({
|
||||
filePath,
|
||||
itemId: 'episode-1',
|
||||
streamIndex: 3,
|
||||
delaySeconds: 1.25,
|
||||
}),
|
||||
true,
|
||||
);
|
||||
|
||||
assert.equal(loadJellyfinSubtitleDelay({ filePath, itemId: 'episode-1', streamIndex: 3 }), 1.25);
|
||||
assert.equal(loadJellyfinSubtitleDelay({ filePath, itemId: 'episode-1', streamIndex: 4 }), null);
|
||||
});
|
||||
|
||||
test('jellyfin subtitle delay store preserves other stream delays when updating one stream', () => {
|
||||
const filePath = statePath('delays.json');
|
||||
|
||||
saveJellyfinSubtitleDelay({ filePath, itemId: 'episode-1', streamIndex: 3, delaySeconds: 1.25 });
|
||||
saveJellyfinSubtitleDelay({ filePath, itemId: 'episode-1', streamIndex: 4, delaySeconds: -0.5 });
|
||||
saveJellyfinSubtitleDelay({ filePath, itemId: 'episode-1', streamIndex: 3, delaySeconds: 2 });
|
||||
|
||||
assert.equal(loadJellyfinSubtitleDelay({ filePath, itemId: 'episode-1', streamIndex: 3 }), 2);
|
||||
assert.equal(loadJellyfinSubtitleDelay({ filePath, itemId: 'episode-1', streamIndex: 4 }), -0.5);
|
||||
});
|
||||
|
||||
test('jellyfin subtitle delay store ignores invalid files and values', () => {
|
||||
const filePath = statePath('delays.json');
|
||||
fs.writeFileSync(filePath, '{');
|
||||
|
||||
assert.equal(loadJellyfinSubtitleDelay({ filePath, itemId: 'episode-1', streamIndex: 3 }), null);
|
||||
assert.equal(
|
||||
saveJellyfinSubtitleDelay({
|
||||
filePath,
|
||||
itemId: 'episode-1',
|
||||
streamIndex: 3,
|
||||
delaySeconds: Number.NaN,
|
||||
}),
|
||||
false,
|
||||
);
|
||||
});
|
||||
@@ -1,66 +0,0 @@
|
||||
import * as fs from 'fs';
|
||||
import * as path from 'path';
|
||||
|
||||
type JellyfinSubtitleDelayStore = {
|
||||
version?: unknown;
|
||||
delays?: unknown;
|
||||
};
|
||||
|
||||
type JellyfinSubtitleDelayParams = {
|
||||
filePath: string;
|
||||
itemId: string;
|
||||
streamIndex: number;
|
||||
};
|
||||
|
||||
type SaveJellyfinSubtitleDelayParams = JellyfinSubtitleDelayParams & {
|
||||
delaySeconds: number;
|
||||
};
|
||||
|
||||
function storeKey(itemId: string, streamIndex: number): string {
|
||||
return JSON.stringify([itemId, streamIndex]);
|
||||
}
|
||||
|
||||
function readDelayMap(filePath: string): Record<string, number> {
|
||||
try {
|
||||
if (!fs.existsSync(filePath)) return {};
|
||||
const parsed = JSON.parse(fs.readFileSync(filePath, 'utf-8')) as JellyfinSubtitleDelayStore;
|
||||
if (
|
||||
!parsed ||
|
||||
typeof parsed !== 'object' ||
|
||||
!parsed.delays ||
|
||||
typeof parsed.delays !== 'object'
|
||||
) {
|
||||
return {};
|
||||
}
|
||||
const delays: Record<string, number> = {};
|
||||
for (const [key, value] of Object.entries(parsed.delays as Record<string, unknown>)) {
|
||||
if (typeof value === 'number' && Number.isFinite(value)) {
|
||||
delays[key] = value;
|
||||
}
|
||||
}
|
||||
return delays;
|
||||
} catch {
|
||||
return {};
|
||||
}
|
||||
}
|
||||
|
||||
export function loadJellyfinSubtitleDelay(params: JellyfinSubtitleDelayParams): number | null {
|
||||
const delay = readDelayMap(params.filePath)[storeKey(params.itemId, params.streamIndex)];
|
||||
return typeof delay === 'number' && Number.isFinite(delay) ? delay : null;
|
||||
}
|
||||
|
||||
export function saveJellyfinSubtitleDelay(params: SaveJellyfinSubtitleDelayParams): boolean {
|
||||
if (!Number.isFinite(params.delaySeconds)) return false;
|
||||
try {
|
||||
const delays = readDelayMap(params.filePath);
|
||||
delays[storeKey(params.itemId, params.streamIndex)] = params.delaySeconds;
|
||||
const dir = path.dirname(params.filePath);
|
||||
if (!fs.existsSync(dir)) {
|
||||
fs.mkdirSync(dir, { recursive: true });
|
||||
}
|
||||
fs.writeFileSync(params.filePath, JSON.stringify({ version: 1, delays }, null, 2));
|
||||
return true;
|
||||
} catch {
|
||||
return false;
|
||||
}
|
||||
}
|
||||
@@ -244,6 +244,35 @@ test('handleMultiCopyDigit copies available history and reports truncation', ()
|
||||
assert.equal(osd.at(-1), 'Only 2 lines available, copied 2');
|
||||
});
|
||||
|
||||
test('handleMultiCopyDigit copies backward from the current subtitle after a backward seek', () => {
|
||||
const copied: string[] = [];
|
||||
const tracker = new SubtitleTimingTracker();
|
||||
|
||||
try {
|
||||
tracker.recordSubtitle('A', 1, 2);
|
||||
tracker.recordSubtitle('B', 3, 4);
|
||||
tracker.recordSubtitle('C', 5, 6);
|
||||
tracker.recordSubtitle('B', 3, 4);
|
||||
|
||||
const deps = {
|
||||
subtitleTimingTracker: tracker,
|
||||
writeClipboardText: (text: string) => copied.push(text),
|
||||
showMpvOsd: () => {},
|
||||
};
|
||||
|
||||
handleMultiCopyDigit(1, deps);
|
||||
handleMultiCopyDigit(2, deps);
|
||||
|
||||
assert.deepEqual(copied, ['B', 'A\n\nB']);
|
||||
assert.deepEqual(tracker.getRecentEntries(2), [
|
||||
{ displayText: 'A', startTime: 1, endTime: 2, secondaryText: undefined },
|
||||
{ displayText: 'B', startTime: 3, endTime: 4, secondaryText: undefined },
|
||||
]);
|
||||
} finally {
|
||||
tracker.destroy();
|
||||
}
|
||||
});
|
||||
|
||||
test('handleMineSentenceDigit reports async create failures', async () => {
|
||||
const osd: string[] = [];
|
||||
const logs: Array<{ message: string; err: unknown }> = [];
|
||||
@@ -344,6 +373,22 @@ test('handleMineSentenceDigit keeps per-entry timings when subtitle text repeats
|
||||
}
|
||||
});
|
||||
|
||||
test('subtitle timing history preserves adjacent repeated text with distinct timings', () => {
|
||||
const tracker = new SubtitleTimingTracker();
|
||||
|
||||
try {
|
||||
tracker.recordSubtitle('same', 1, 2);
|
||||
tracker.recordSubtitle('same', 3, 4);
|
||||
|
||||
assert.deepEqual(tracker.getRecentEntries(2), [
|
||||
{ displayText: 'same', startTime: 1, endTime: 2, secondaryText: undefined },
|
||||
{ displayText: 'same', startTime: 3, endTime: 4, secondaryText: undefined },
|
||||
]);
|
||||
} finally {
|
||||
tracker.destroy();
|
||||
}
|
||||
});
|
||||
|
||||
test('handleMineSentenceDigit joins per-entry secondary subtitles when available', async () => {
|
||||
const created: Array<{ sentence: string; secondarySub?: string }> = [];
|
||||
const tracker = new SubtitleTimingTracker();
|
||||
|
||||
@@ -83,6 +83,7 @@ function createDeps(overrides: Partial<MpvProtocolHandleMessageDeps> = {}): {
|
||||
state.secondarySubText = text;
|
||||
},
|
||||
resolvePendingRequest: () => false,
|
||||
shouldEnforceSecondarySubVisibilityHidden: () => true,
|
||||
setSecondarySubVisibility: () => {},
|
||||
syncCurrentAudioStreamIndex: () => {},
|
||||
setCurrentAudioTrackId: () => {},
|
||||
@@ -198,6 +199,21 @@ test('dispatchMpvProtocolMessage rejects decimal subtitle track IDs', async () =
|
||||
assert.deepEqual(state.events, [{ sid: null }, { sid: null }, { sid: null }, { sid: null }]);
|
||||
});
|
||||
|
||||
test('dispatchMpvProtocolMessage hides native secondary subtitles after a track change', async () => {
|
||||
const visibilityChanges: boolean[] = [];
|
||||
const { deps, state } = createDeps({
|
||||
setSecondarySubVisibility: (visible) => visibilityChanges.push(visible),
|
||||
});
|
||||
|
||||
await dispatchMpvProtocolMessage(
|
||||
{ event: 'property-change', name: 'secondary-sid', data: '4' },
|
||||
deps,
|
||||
);
|
||||
|
||||
assert.deepEqual(visibilityChanges, [false]);
|
||||
assert.deepEqual(state.events, [{ sid: 4 }]);
|
||||
});
|
||||
|
||||
test('dispatchMpvProtocolMessage enforces sub-visibility hidden when overlay suppression is enabled', async () => {
|
||||
const { deps, state } = createDeps({
|
||||
isVisibleOverlayVisible: () => true,
|
||||
@@ -239,6 +255,24 @@ test('dispatchMpvProtocolMessage skips sub-visibility suppression when overlay i
|
||||
assert.equal(state.commands.length, 0);
|
||||
});
|
||||
|
||||
test('dispatchMpvProtocolMessage corrects native secondary subtitle visibility', async () => {
|
||||
const visibilityChanges: boolean[] = [];
|
||||
const { deps } = createDeps({
|
||||
setSecondarySubVisibility: (visible) => visibilityChanges.push(visible),
|
||||
});
|
||||
|
||||
await dispatchMpvProtocolMessage(
|
||||
{ event: 'property-change', name: 'secondary-sub-visibility', data: 'yes' },
|
||||
deps,
|
||||
);
|
||||
await dispatchMpvProtocolMessage(
|
||||
{ event: 'property-change', name: 'secondary-sub-visibility', data: 'no' },
|
||||
deps,
|
||||
);
|
||||
|
||||
assert.deepEqual(visibilityChanges, [false]);
|
||||
});
|
||||
|
||||
test('dispatchMpvProtocolMessage sets secondary subtitle track based on track list response', async () => {
|
||||
const { deps, state } = createDeps();
|
||||
|
||||
|
||||
@@ -72,6 +72,7 @@ export interface MpvProtocolHandleMessageDeps {
|
||||
emitSubtitleMetricsChange: (payload: Partial<MpvSubtitleRenderMetrics>) => void;
|
||||
setCurrentSecondarySubText: (text: string) => void;
|
||||
resolvePendingRequest: (requestId: number, message: MpvMessage) => boolean;
|
||||
shouldEnforceSecondarySubVisibilityHidden: () => boolean;
|
||||
setSecondarySubVisibility: (visible: boolean) => void;
|
||||
syncCurrentAudioStreamIndex: () => void;
|
||||
setCurrentAudioTrackId: (value: number | null) => void;
|
||||
@@ -285,6 +286,9 @@ export async function dispatchMpvProtocolMessage(
|
||||
: null;
|
||||
deps.emitSubtitleTrackChange({ sid: sid !== null && Number.isInteger(sid) ? sid : null });
|
||||
} else if (msg.name === 'secondary-sid') {
|
||||
if (deps.shouldEnforceSecondarySubVisibilityHidden()) {
|
||||
deps.setSecondarySubVisibility(false);
|
||||
}
|
||||
const sid =
|
||||
typeof msg.data === 'number'
|
||||
? msg.data
|
||||
@@ -375,6 +379,11 @@ export async function dispatchMpvProtocolMessage(
|
||||
if (deps.isVisibleOverlayVisible() && asBoolean(msg.data, false)) {
|
||||
deps.sendCommand({ command: ['set_property', 'sub-visibility', false] });
|
||||
}
|
||||
} else if (msg.name === 'secondary-sub-visibility') {
|
||||
const visible = parseVisibilityProperty(msg.data);
|
||||
if (deps.shouldEnforceSecondarySubVisibilityHidden() && visible === true) {
|
||||
deps.setSecondarySubVisibility(false);
|
||||
}
|
||||
} else if (msg.name === 'sub-use-margins') {
|
||||
deps.emitSubtitleMetricsChange({
|
||||
subUseMargins: asBoolean(msg.data, deps.getSubtitleMetrics().subUseMargins),
|
||||
|
||||
@@ -652,7 +652,7 @@ test('MpvIpcClient captures and disables secondary subtitle visibility on reques
|
||||
]);
|
||||
});
|
||||
|
||||
test('MpvIpcClient restorePreviousSecondarySubVisibility restores and clears tracked value', async () => {
|
||||
test('MpvIpcClient restores secondary subtitle visibility and relinquishes suppression', async () => {
|
||||
const commands: unknown[] = [];
|
||||
const client = new MpvIpcClient('/tmp/mpv.sock', makeDeps());
|
||||
const previous: boolean[] = [];
|
||||
@@ -671,6 +671,12 @@ test('MpvIpcClient restorePreviousSecondarySubVisibility restores and clears tra
|
||||
});
|
||||
client.restorePreviousSecondarySubVisibility();
|
||||
|
||||
await invokeHandleMessage(client, {
|
||||
event: 'property-change',
|
||||
name: 'secondary-sub-visibility',
|
||||
data: 'yes',
|
||||
});
|
||||
|
||||
assert.equal(previous[0], true);
|
||||
assert.equal(previous.length, 1);
|
||||
assert.deepEqual(commands, [
|
||||
@@ -682,8 +688,53 @@ test('MpvIpcClient restorePreviousSecondarySubVisibility restores and clears tra
|
||||
},
|
||||
]);
|
||||
|
||||
await invokeHandleMessage(client, {
|
||||
event: 'property-change',
|
||||
name: 'secondary-sub-visibility',
|
||||
data: 'yes',
|
||||
});
|
||||
assert.equal(commands.length, 2);
|
||||
|
||||
client.restorePreviousSecondarySubVisibility();
|
||||
assert.equal(commands.length, 2);
|
||||
|
||||
const callbacks = (client as any).transport.callbacks;
|
||||
callbacks.onConnect();
|
||||
commands.length = 0;
|
||||
|
||||
await invokeHandleMessage(client, {
|
||||
event: 'property-change',
|
||||
name: 'secondary-sub-visibility',
|
||||
data: 'yes',
|
||||
});
|
||||
assert.deepEqual(commands, [{ command: ['set_property', 'secondary-sub-visibility', 'no'] }]);
|
||||
});
|
||||
|
||||
test('MpvIpcClient keeps secondary subtitle suppression when restoration send fails', async () => {
|
||||
const commands: unknown[] = [];
|
||||
const client = new MpvIpcClient('/tmp/mpv.sock', makeDeps());
|
||||
|
||||
(client as any).send = (payload: unknown) => {
|
||||
commands.push(payload);
|
||||
return false;
|
||||
};
|
||||
|
||||
await invokeHandleMessage(client, {
|
||||
request_id: MPV_REQUEST_ID_SECONDARY_SUB_VISIBILITY,
|
||||
data: 'yes',
|
||||
});
|
||||
client.restorePreviousSecondarySubVisibility();
|
||||
await invokeHandleMessage(client, {
|
||||
event: 'property-change',
|
||||
name: 'secondary-sid',
|
||||
data: 4,
|
||||
});
|
||||
|
||||
assert.deepEqual(commands, [
|
||||
{ command: ['set_property', 'secondary-sub-visibility', 'no'] },
|
||||
{ command: ['set_property', 'secondary-sub-visibility', 'yes'] },
|
||||
{ command: ['set_property', 'secondary-sub-visibility', 'no'] },
|
||||
]);
|
||||
});
|
||||
|
||||
test('MpvIpcClient updates current audio stream index from track list', async () => {
|
||||
|
||||
@@ -184,6 +184,7 @@ export class MpvIpcClient implements MpvClient {
|
||||
osdDimensions: null,
|
||||
};
|
||||
private previousSecondarySubVisibility: boolean | null = null;
|
||||
private enforceSecondarySubVisibilityHidden = true;
|
||||
private playbackPaused: boolean | null = null;
|
||||
private pauseAtTime: number | null = null;
|
||||
private pendingPauseAtSubEnd = false;
|
||||
@@ -199,6 +200,7 @@ export class MpvIpcClient implements MpvClient {
|
||||
socketFactory: deps.socketFactory,
|
||||
connectTimeoutMs: deps.connectTimeoutMs,
|
||||
onConnect: () => {
|
||||
this.enforceSecondarySubVisibilityHidden = true;
|
||||
this.connected = true;
|
||||
this.connecting = false;
|
||||
this.socket = this.transport.getSocket();
|
||||
@@ -476,6 +478,7 @@ export class MpvIpcClient implements MpvClient {
|
||||
},
|
||||
resolvePendingRequest: (requestId: number, message: MpvMessage) =>
|
||||
this.tryResolvePendingRequest(requestId, message),
|
||||
shouldEnforceSecondarySubVisibilityHidden: () => this.enforceSecondarySubVisibilityHidden,
|
||||
setSecondarySubVisibility: (visible: boolean) => this.setSecondarySubVisibility(visible),
|
||||
syncCurrentAudioStreamIndex: () => {
|
||||
this.syncCurrentAudioStreamIndex();
|
||||
@@ -647,9 +650,11 @@ export class MpvIpcClient implements MpvClient {
|
||||
restorePreviousSecondarySubVisibility(): void {
|
||||
const previous = this.previousSecondarySubVisibility;
|
||||
if (previous === null) return;
|
||||
this.send({
|
||||
const restored = this.send({
|
||||
command: ['set_property', 'secondary-sub-visibility', previous ? 'yes' : 'no'],
|
||||
});
|
||||
if (!restored) return;
|
||||
this.enforceSecondarySubVisibilityHidden = false;
|
||||
this.previousSecondarySubVisibility = null;
|
||||
}
|
||||
|
||||
|
||||
@@ -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[1]?.assFurigana, ['ともり']);
|
||||
assert.deepEqual(cues[0]?.assFurigana, ['たき', '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',
|
||||
'隠し貯蔵のミルクまんじゅう➡',
|
||||
'（山田の母）ほんなら\n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[4]?.assFurigana, ['ぜったい ちが']);
|
||||
assert.deepEqual(cues[2]?.assFurigana, ['かく', 'ちょぞう']);
|
||||
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', () => {
|
||||
|
||||
@@ -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[] {
|
||||
|
||||
@@ -1,73 +0,0 @@
|
||||
import assert from 'node:assert/strict';
|
||||
import test from 'node:test';
|
||||
import { estimateSubtitleTimingOffset } from './subtitle-timing-offset';
|
||||
|
||||
function cue(startTime: number) {
|
||||
return { startTime, endTime: startTime + 1, text: `cue ${startTime}` };
|
||||
}
|
||||
|
||||
test('estimate subtitle timing offset detects a late Jellyfin subtitle timeline', () => {
|
||||
const primary = [
|
||||
34.935, 36.937, 41.441, 45.279, 48.115, 52.286, 54.955, 59.793, 63.63, 67.634, 76.643, 80.814,
|
||||
87.988, 90.991, 94.094, 97.097,
|
||||
].map(cue);
|
||||
const reference = [
|
||||
3.46, 9.48, 13.61, 21.4, 28.16, 32.06, 35.93, 45.1, 56.57, 59.68, 62.44, 65.56,
|
||||
].map(cue);
|
||||
|
||||
const result = estimateSubtitleTimingOffset(primary, reference);
|
||||
|
||||
assert.ok(result);
|
||||
assert.ok(result.offsetSeconds > -32);
|
||||
assert.ok(result.offsetSeconds < -31);
|
||||
assert.ok(result.matchCount >= 8);
|
||||
assert.ok(result.meanErrorSeconds <= 0.75);
|
||||
});
|
||||
|
||||
test('estimate subtitle timing offset favors the early episode timeline', () => {
|
||||
const primary = [
|
||||
34.935, 36.937, 41.441, 45.279, 48.115, 52.286, 54.955, 59.793, 63.63, 67.634, 76.643, 80.814,
|
||||
87.988, 90.991, 94.094, 97.097, 207.974, 212.579, 222.422, 228.095, 232.432, 238.271, 244.778,
|
||||
246.78, 249.282, 251.284, 253.62, 256.289, 259.626, 262.129, 264.965, 267.634, 270.303, 274.407,
|
||||
277.077, 280.08, 284.084, 288.421, 291.925, 295.262, 298.431, 301.101, 306.773, 308.942,
|
||||
312.946, 316.283, 321.621, 326.626, 331.131, 336.069, 340.407, 343.41, 351.418, 355.422,
|
||||
357.924, 362.429, 365.432, 370.604, 373.273, 377.944, 381.114, 384.618, 387.621, 390.957,
|
||||
396.73, 399.232, 401.568, 403.57, 405.572, 407.574, 409.743, 412.746, 418.752, 425.258, 427.26,
|
||||
435.602, 440.44, 442.942, 445.445, 449.783,
|
||||
].map(cue);
|
||||
const reference = [
|
||||
3.46, 9.48, 13.61, 21.4, 28.16, 32.06, 35.93, 45.1, 56.57, 59.68, 62.44, 65.56, 165.77, 172.81,
|
||||
176.1, 177.27, 186.33, 191.33, 195.78, 201.83, 212.9, 214.09, 216.73, 220.2, 222.91, 225.65,
|
||||
232.8, 237.92, 242.23, 243.28, 247.53, 252.04, 255.9, 258.86, 262.09, 264.43, 276.07, 278.01,
|
||||
280.98, 285.67, 289.89, 294.57, 300, 303.56, 308.58, 316.37, 318.38, 319.86, 325.38, 328.82,
|
||||
333.68, 335.26, 336.82, 340.11, 342.11, 344.36, 346.39, 347.53, 350.92, 370.18, 372.88, 376.43,
|
||||
388.2, 390.57, 403.96, 406.36, 409.72, 413.78, 425.55, 432.76, 435.03, 438.06, 443.73, 448.31,
|
||||
450.57, 457.62, 463.41, 465.85, 473.79, 480.59,
|
||||
].map(cue);
|
||||
|
||||
const result = estimateSubtitleTimingOffset(primary, reference);
|
||||
|
||||
assert.ok(result);
|
||||
assert.ok(result.offsetSeconds > -32);
|
||||
assert.ok(result.offsetSeconds < -31);
|
||||
});
|
||||
|
||||
test('estimate subtitle timing offset ignores subtitle timelines that are already aligned', () => {
|
||||
const starts = [1, 5, 9, 14, 20, 25, 31, 38];
|
||||
|
||||
const result = estimateSubtitleTimingOffset(
|
||||
starts.map(cue),
|
||||
starts.map((start) => cue(start + 0.04)),
|
||||
);
|
||||
|
||||
assert.equal(result, null);
|
||||
});
|
||||
|
||||
test('estimate subtitle timing offset rejects weak timeline matches', () => {
|
||||
const primary = [10, 20, 30, 40, 50, 60, 70, 80].map(cue);
|
||||
const reference = [1, 2, 3, 4, 5, 6, 7, 8].map(cue);
|
||||
|
||||
const result = estimateSubtitleTimingOffset(primary, reference);
|
||||
|
||||
assert.equal(result, null);
|
||||
});
|
||||
@@ -1,153 +0,0 @@
|
||||
import type { SubtitleCue } from './subtitle-cue-parser';
|
||||
|
||||
export type SubtitleTimingOffsetResult = {
|
||||
offsetSeconds: number;
|
||||
matchCount: number;
|
||||
meanErrorSeconds: number;
|
||||
maxErrorSeconds: number;
|
||||
};
|
||||
|
||||
export type SubtitleTimingOffsetOptions = {
|
||||
maxCueCount?: number;
|
||||
maxOffsetSeconds?: number;
|
||||
matchThresholdSeconds?: number;
|
||||
maxMeanErrorSeconds?: number;
|
||||
minMatchCount?: number;
|
||||
minMatchRatio?: number;
|
||||
minUsefulOffsetSeconds?: number;
|
||||
};
|
||||
|
||||
type OffsetScore = SubtitleTimingOffsetResult;
|
||||
|
||||
const DEFAULT_MAX_CUE_COUNT = 60;
|
||||
const DEFAULT_MAX_OFFSET_SECONDS = 180;
|
||||
const DEFAULT_MATCH_THRESHOLD_SECONDS = 1;
|
||||
const DEFAULT_MAX_MEAN_ERROR_SECONDS = 0.75;
|
||||
const DEFAULT_MIN_MATCH_COUNT = 8;
|
||||
const DEFAULT_MIN_MATCH_RATIO = 0.25;
|
||||
const DEFAULT_MIN_USEFUL_OFFSET_SECONDS = 0.25;
|
||||
|
||||
function normalizeCueStarts(cues: SubtitleCue[], maxCueCount: number): number[] {
|
||||
const starts = cues
|
||||
.map((cue) => cue.startTime)
|
||||
.filter((start) => Number.isFinite(start) && start >= 0)
|
||||
.sort((a, b) => a - b);
|
||||
const deduped: number[] = [];
|
||||
for (const start of starts) {
|
||||
const previous = deduped[deduped.length - 1];
|
||||
if (previous === undefined || Math.abs(start - previous) > 0.05) {
|
||||
deduped.push(start);
|
||||
}
|
||||
if (deduped.length >= maxCueCount) {
|
||||
break;
|
||||
}
|
||||
}
|
||||
return deduped;
|
||||
}
|
||||
|
||||
function roundToMillis(value: number): number {
|
||||
return Math.round(value * 1000) / 1000;
|
||||
}
|
||||
|
||||
function scoreOffset(
|
||||
primaryStarts: number[],
|
||||
referenceStarts: number[],
|
||||
offsetSeconds: number,
|
||||
matchThresholdSeconds: number,
|
||||
): OffsetScore {
|
||||
let primaryIndex = 0;
|
||||
let referenceIndex = 0;
|
||||
let matchCount = 0;
|
||||
let totalErrorSeconds = 0;
|
||||
let maxErrorSeconds = 0;
|
||||
|
||||
while (primaryIndex < primaryStarts.length && referenceIndex < referenceStarts.length) {
|
||||
const shiftedPrimary = primaryStarts[primaryIndex]! + offsetSeconds;
|
||||
const reference = referenceStarts[referenceIndex]!;
|
||||
const errorSeconds = Math.abs(shiftedPrimary - reference);
|
||||
if (errorSeconds <= matchThresholdSeconds) {
|
||||
matchCount += 1;
|
||||
totalErrorSeconds += errorSeconds;
|
||||
maxErrorSeconds = Math.max(maxErrorSeconds, errorSeconds);
|
||||
primaryIndex += 1;
|
||||
referenceIndex += 1;
|
||||
continue;
|
||||
}
|
||||
|
||||
if (shiftedPrimary < reference) {
|
||||
primaryIndex += 1;
|
||||
} else {
|
||||
referenceIndex += 1;
|
||||
}
|
||||
}
|
||||
|
||||
return {
|
||||
offsetSeconds,
|
||||
matchCount,
|
||||
meanErrorSeconds: matchCount > 0 ? totalErrorSeconds / matchCount : Number.POSITIVE_INFINITY,
|
||||
maxErrorSeconds,
|
||||
};
|
||||
}
|
||||
|
||||
function isBetterScore(next: OffsetScore, current: OffsetScore | null): boolean {
|
||||
if (current === null) return true;
|
||||
if (next.matchCount !== current.matchCount) return next.matchCount > current.matchCount;
|
||||
if (next.meanErrorSeconds !== current.meanErrorSeconds) {
|
||||
return next.meanErrorSeconds < current.meanErrorSeconds;
|
||||
}
|
||||
return Math.abs(next.offsetSeconds) < Math.abs(current.offsetSeconds);
|
||||
}
|
||||
|
||||
export function estimateSubtitleTimingOffset(
|
||||
primaryCues: SubtitleCue[],
|
||||
referenceCues: SubtitleCue[],
|
||||
options: SubtitleTimingOffsetOptions = {},
|
||||
): SubtitleTimingOffsetResult | null {
|
||||
const maxCueCount = options.maxCueCount ?? DEFAULT_MAX_CUE_COUNT;
|
||||
const maxOffsetSeconds = options.maxOffsetSeconds ?? DEFAULT_MAX_OFFSET_SECONDS;
|
||||
const matchThresholdSeconds = options.matchThresholdSeconds ?? DEFAULT_MATCH_THRESHOLD_SECONDS;
|
||||
const maxMeanErrorSeconds = options.maxMeanErrorSeconds ?? DEFAULT_MAX_MEAN_ERROR_SECONDS;
|
||||
const minMatchCount = options.minMatchCount ?? DEFAULT_MIN_MATCH_COUNT;
|
||||
const minMatchRatio = options.minMatchRatio ?? DEFAULT_MIN_MATCH_RATIO;
|
||||
const minUsefulOffsetSeconds =
|
||||
options.minUsefulOffsetSeconds ?? DEFAULT_MIN_USEFUL_OFFSET_SECONDS;
|
||||
|
||||
const primaryStarts = normalizeCueStarts(primaryCues, maxCueCount);
|
||||
const referenceStarts = normalizeCueStarts(referenceCues, maxCueCount);
|
||||
const comparableCueCount = Math.min(primaryStarts.length, referenceStarts.length);
|
||||
if (comparableCueCount < minMatchCount) {
|
||||
return null;
|
||||
}
|
||||
|
||||
const candidates = new Set<number>();
|
||||
for (const primaryStart of primaryStarts) {
|
||||
for (const referenceStart of referenceStarts) {
|
||||
const offsetSeconds = roundToMillis(referenceStart - primaryStart);
|
||||
if (Math.abs(offsetSeconds) <= maxOffsetSeconds) {
|
||||
candidates.add(offsetSeconds);
|
||||
}
|
||||
}
|
||||
}
|
||||
|
||||
let best: OffsetScore | null = null;
|
||||
for (const offsetSeconds of candidates) {
|
||||
if (Math.abs(offsetSeconds) < minUsefulOffsetSeconds) {
|
||||
continue;
|
||||
}
|
||||
const score = scoreOffset(primaryStarts, referenceStarts, offsetSeconds, matchThresholdSeconds);
|
||||
if (score.matchCount < minMatchCount) {
|
||||
continue;
|
||||
}
|
||||
if (score.matchCount / comparableCueCount < minMatchRatio) {
|
||||
continue;
|
||||
}
|
||||
if (score.meanErrorSeconds > maxMeanErrorSeconds) {
|
||||
continue;
|
||||
}
|
||||
if (isBetterScore(score, best)) {
|
||||
best = score;
|
||||
}
|
||||
}
|
||||
|
||||
return best;
|
||||
}
|
||||
Reference in New Issue
Block a user