align-captions

Align existing script text to audio for karaoke-style captions. Uses Qwen3-ForcedAligner-0.6B (~30ms precision) and jieba for Chinese word segmentation (groups characters into natural words).

Pipeline

Script + Audio
    ↓
Qwen3-ForcedAligner (character-level timestamps)
    ↓
Jieba word segmentation (characters → Chinese words)
    ↓
Position-based phrase matching (words → phrases)
    ↓
Output: phrases with embedded word timestamps

Usage