Re-Transcribe Video Without Losing Edits

Re-Transcribe Video Without Losing Edits

Changed your mind about the model? Need to switch languages? Re-run Whisper at any time. Velocaption’s merge algorithm preserves every manual correction you’ve made: edited words are matched by timestamp and kept in place.

How It Works

1. Open the transcription panel: navigate to the same transcription UI you used for the initial transcription.

2. Change your settings: select a different model (e.g., switch from small to medium for better accuracy) or change the language setting. Any transcription setting can be modified.

3. Click Transcribe: Whisper runs again on the full audio. This produces a completely new set of word-level timestamps.

4. Automatic merge: when the new transcription completes, the merge algorithm runs:
   • Each new word is compared to existing edited words by timestamp (0.3 second tolerance)
   • If a new word overlaps with an edited word, the edited word wins: new output is discarded
   • Unedited words are replaced with the new transcription
   • Orphaned edited words (no matching new word nearby) are preserved and sorted chronologically

5. Review the result: the merged transcript appears in the sidebar. Edited words keep their flags and original text. New words may have better accuracy from the upgraded model.

Key Details

Timestamp tolerance is 0.3 seconds. A new word is considered a “match” for an existing edited word if their timestamps are within 300ms. This accounts for minor timing shifts between Whisper runs.

Segment IDs are preserved. After merge, segments are re-grouped and matched to their previous IDs using the first word’s start time. This preserves per-phrase position overrides and other segment-level metadata.

Orphaned edits are never lost. If you edited a word and the new transcription doesn’t produce anything near that timestamp, the edited word is still kept and inserted in chronological order.

What’s Next

• Local Whisper Transcription: model selection and transcription settings
• Edit Words in the Transcript: confidence-guided word editing
• Find and Replace Across Transcript: batch-correct repeated errors