Fix and verify karaoke sync upgrade once

Fix and reverify karaoke sync upgrade #1

Workflow file for this run

.github/workflows/karaoke-sync-fixup-once.yml at 85040fa

	name: Fix and verify karaoke sync upgrade once

	on:
	push:
	branches: [fix/karaoke-preserve-word-timing]
	paths: [".github/workflows/karaoke-sync-fixup-once.yml"]

	permissions:
	contents: write

	jobs:
	apply-fix-and-verify:
	runs-on: ubuntu-latest
	timeout-minutes: 35
	steps:
	- name: Check out PR branch
	uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
	with:
	ref: fix/karaoke-preserve-word-timing
	persist-credentials: true

	- name: Set up Java 17
	uses: actions/setup-java@cf277c60eb25467037889841efdb72551f06f6c3 # v4
	with:
	distribution: temurin
	java-version: "17"

	- name: Set up Android SDK
	uses: android-actions/setup-android@9fc6c4e9069bf8d3d10b2204b1fb8f6ef7065407 # v3

	- name: Install Android build SDK
	run: sdkmanager "platforms;android-36" "build-tools;36.0.0"

	- name: Apply original options 1-4 upgrade
	shell: bash
	run: \|
	python - <<'PY'
	from pathlib import Path
	import base64, gzip, re, subprocess
	workflow = Path('.github/workflows/karaoke-sync-upgrade-once.yml').read_text()
	match = re.search(r"printf '%s' '([A-Za-z0-9+/=]+)' \\\| base64", workflow)
	if not match:
	raise SystemExit('Could not recover the original upgrade payload')
	script = gzip.decompress(base64.b64decode(match.group(1)))
	path = Path('/tmp/apply-karaoke-sync.py')
	path.write_bytes(script)
	subprocess.run(['python', str(path)], check=True)
	PY

	- name: Fix timing model and estimator
	shell: bash
	run: \|
	cat > app/src/main/java/com/kienhoang/dualsubreplay/data/SubtitleModels.kt <<'EOF'
	package com.kienhoang.dualsubreplay.data

	data class RawCaptionCue(
	val startMs: Long,
	val endMs: Long,
	val text: String,
	val words: List<SubtitleWord> = emptyList(),
	)

	data class SubtitleSegment(
	val id: Long,
	val startMs: Long,
	val endMs: Long,
	val originalText: String,
	val translatedText: String? = null,
	val words: List<SubtitleWord> = emptyList(),
	)

	enum class SubtitleTimingSource {
	YOUTUBE_EXACT,
	YOUTUBE_DOM_OBSERVED,
	ESTIMATED,
	}

	/** A single spoken word/chunk with its absolute timing inside the video. */
	data class SubtitleWord(
	val text: String,
	val startMs: Long,
	val endMs: Long,
	val timingSource: SubtitleTimingSource = SubtitleTimingSource.ESTIMATED,
	)

	data class CaptionTrackResult(
	val languageCode: String,
	val isGenerated: Boolean,
	val cues: List<RawCaptionCue>,
	val availableLanguages: List<CaptionLanguage> = emptyList(),
	)

	data class CaptionLanguage(
	val code: String,
	val name: String,
	)
	EOF

	cat > app/src/main/java/com/kienhoang/dualsubreplay/data/WordTiming.kt <<'EOF'
	package com.kienhoang.dualsubreplay.data

	/**
	* Word-level timing helpers shared by the caption parser, the merger, and the
	* real-time spoken-word highlight in the UI.
	*/

	/**
	* With a frame-synchronised WebView clock most polling latency disappears.
	* Keep only a tiny render lead so Compose can paint the next word without
	* visibly jumping ahead of the speaker.
	*/
	internal const val KARAOKE_HIGHLIGHT_LEAD_MS = 20L
	private const val MIN_ESTIMATED_WORD_MS = 60L
	private val sentencePause = Regex("[.!?。！？…]+[\\\"'’”)]*$")
	private val clausePause = Regex("[,;:，；：]+[\\\"'’”)]*$")
	private val latinVowelGroups = Regex("(?i)[aeiouy]+")

	/**
	* A speech-oriented fallback weight. It avoids giving very long written words
	* an unrealistically huge share of a cue and reserves a little time for
	* punctuation pauses. This is still explicitly ESTIMATED timing.
	*/
	private fun estimatedTimingWeight(token: String): Long {
	val spokenCharacters = token.count(Char::isLetterOrDigit).coerceAtLeast(1)
	val vowelGroups = latinVowelGroups.findAll(token).count()
	val roughSyllables = maxOf(vowelGroups, (spokenCharacters + 2) / 3)
	.coerceIn(1, 6)
	val pauseWeight = when {
	sentencePause.containsMatchIn(token) -> 3
	clausePause.containsMatchIn(token) -> 2
	else -> 0
	}
	return (roughSyllables + pauseWeight).toLong()
	}

	/**
	* Splits text into words and estimates their boundaries from speech-oriented
	* weights. When the cue is long enough, every word receives a small minimum
	* slice before the remaining duration is distributed by the weights.
	*/
	internal fun estimateWordTimings(text: String, startMs: Long, endMs: Long): List<SubtitleWord> {
	val tokens = text.split(Regex("\\s+")).filter(String::isNotBlank)
	if (tokens.isEmpty() \|\| endMs <= startMs) return emptyList()

	val duration = endMs - startMs
	val minimumPerWord = if (duration >= tokens.size * MIN_ESTIMATED_WORD_MS) {
	MIN_ESTIMATED_WORD_MS
	} else {
	0L
	}
	val reserved = minimumPerWord * tokens.size
	val distributable = (duration - reserved).coerceAtLeast(0L)
	val weights = tokens.map(::estimatedTimingWeight)
	val totalWeight = weights.sum().coerceAtLeast(1L)

	var consumedWeight = 0L
	var cursor = startMs
	return tokens.mapIndexed { index, token ->
	consumedWeight += weights[index]
	val proportionalEnd = if (index == tokens.lastIndex) {
	endMs
	} else {
	startMs +
	minimumPerWord * (index + 1L) +
	distributable * consumedWeight / totalWeight
	}
	val safeEnd = proportionalEnd.coerceIn(cursor, endMs)
	val word = SubtitleWord(
	text = token,
	startMs = cursor,
	endMs = safeEnd,
	timingSource = SubtitleTimingSource.ESTIMATED,
	)
	cursor = safeEnd
	word
	}
	}

	/**
	* Index of the word being spoken at [timeMs]. Between words the previously
	* started word stays highlighted so short gaps do not flicker.
	*/
	internal fun activeWordIndex(words: List<SubtitleWord>, timeMs: Long): Int {
	val firstWord = words.firstOrNull() ?: return -1
	if (timeMs < firstWord.startMs) return -1

	val syncTimeMs = timeMs + KARAOKE_HIGHLIGHT_LEAD_MS
	var result = 0
	for (index in 1 until words.size) {
	if (words[index].startMs <= syncTimeMs) result = index else break
	}
	return result
	}
	EOF

	- name: Test, lint, and build
	run: \|
	set -euo pipefail
	bash ./gradlew testDebugUnitTest lintDebug assembleDebug assembleDebugAndroidTest --stacktrace

	- name: Commit verified upgrade
	shell: bash
	run: \|
	set -euo pipefail
	git config user.name "github-actions[bot]"
	git config user.email "41898282+github-actions[bot]@users.noreply.github.com"
	git add app
	git commit -m "Upgrade karaoke sync with frame clock and DOM hints"
	git push origin HEAD:fix/karaoke-preserve-word-timing

Navigation Menu

Fix and reverify karaoke sync upgrade #1

Workflow file

Fix and reverify karaoke sync upgrade #1

Uh oh!

Workflow file for this run