Repository navigation
FuseClip 0.5.0
FuseClip v0.5.0
This release focuses on fixing caption timing accuracy and word synchronization.
Fixes & Improvements
Accurate Caption & Word-Level Timing
- Dynamic Time Warping (DTW) alignment: Transcription now uses whisper.cpp's DTW token timestamps (
--dtw) instead of default heuristic timestamps, eliminating timing drift where word highlights lagged behind or drifted from spoken audio. - Precise word boundaries: Word boundaries are now resolved against the next token's onset, ensuring each word has an accurate start and end timestamp.
- Snappier visual highlight: Word highlights lead audio onset slightly (80ms) to synchronize naturally with spoken speech and visual perception.
- Graceful fallback: Automatically falls back to standard timestamps if an installed model or binary does not support DTW presets.