Add a full TTML (Timed Text Markup Language) sidecar lyrics parser that extracts
word/syllable-level timing from <span> elements, plus translation and pronunciation
(transliteration) tracks from Apple Music TTML metadata sections.
Backend changes:
- TTML parser (core/lyrics/ttml.go) with support for all TTML time formats,
nested timing contexts, and bare decimal second offsets
- Translation/pronunciation tracks resolved via key-based metadata linking
- Line timing hydration from token-level start/end values
- 'kind' field added to Lyrics model and StructuredLyric API response
(main/translation/pronunciation)
- 'tokenLine' array in API response for word-level timing data
- UTF-8 BOM and UTF-16 LE encoding support for TTML files
- Fix for ambiguous time resolution in pronunciation spans (pre-1-minute)
Frontend changes:
- KaraokeLyricsOverlay rewritten with scrollable multi-line layout,
word-level wipe highlighting with eased alpha transitions,
rAF-driven playback clock with drift correction
- Inline translation (above) and pronunciation (below) each main line,
with smart filtering to hide redundant lines (same normalized text)
- TR/PR toggle buttons and layer selection via selectLyricLayers()
- Click-to-seek: click any lyric line to jump to that position
- Customization popover with font-size sliders and color presets
for each line type (TR/Default/PR), persisted to localStorage
- Smooth font-size transition between active and inactive lines
- Resizable overlay height via drag handle
- lyrics.js: resolveKaraokeTokenWindow, buildSyntheticWordTokens,
findLayerLineIndexForMain, token sorting, collapsed timing detection
API extension (non-breaking, additive):
- tokenLine[].token[] provides per-word start/end timing (ms)
- tokenLine[].index maps back to the corresponding line[] entry
- kind field: 'main', 'translation', 'pronunciation'
- Clients ignoring tokenLine/kind continue to work unchanged
