Edit Generated Subtitles and Check Confidence

Edit Generated Subtitles and Check Confidence

Every word Whisper produces is individually editable. Confidence underlines tell you exactly where to focus your corrections, and every edit is non-destructive: you can always revert to the original.

How It Works

1. Spot errors with confidence underlines: Whisper assigns a confidence score (0.0 to 1.0) to every word. Words are underlined by confidence: yellow (0.7+, probably correct), orange (0.5+, worth checking), red (below 0.5, likely wrong). Words with high confidence have no underline.

2. Double-click to edit: click any word in the transcript sidebar to select it, double-click to enter edit mode. The word becomes a text input. Type your correction.

3. Confirm the edit: press Enter or click outside the word to save. The word is flagged as _edited and the original text is preserved for revert. Saves are debounced (150ms) so rapid edits don’t cause performance issues.

4. Right-click for more options: right-click any word to open the context menu. It shows the exact confidence percentage, a “Revert to original” option (for edited words), and “Replace all occurrences” for batch corrections.

5. Revert if needed: changed your mind? Right-click the edited word and select “Revert to original” to restore Whisper’s original text. The edit flag is removed and the original confidence score is preserved.

Key Details

Focus on red and orange words. You don’t need to review every word. Confidence underlines guide you to the ~5-10% of words that Whisper struggled with: names, technical terms, and mumbled speech.

Edits are O(1) fast-path. Word text changes update only the segment entity map. The word list, segment IDs, and timeline intervals are unchanged: so the virtualizer and timeline don’t re-render.

Original confidence is preserved. Even after editing a word, Whisper’s original confidence score stays attached. This means confidence underlines reflect transcription difficulty, not whether you’ve already fixed it.

What’s Next

• Find and Replace Across Transcript: batch-correct a word Whisper consistently misspells
• Split and Merge Words: fix timing issues by splitting or combining words
• Re-Transcribe Without Losing Edits: run a better model while keeping corrections