
Word-Level Editing
Click Any Word — Jump to That Exact Moment in the Video
Traditional subtitle editors work on whole blocks. CutSnap goes deeper: every word in your transcript carries its own timestamp, so you can click a single word, see the video jump to that millisecond, and fix it in context. It is the fastest way to correct AI transcription errors and fine-tune subtitle timing.
How CutSnap generates word timestamps
CutSnap's transcription pipeline outputs two levels of timestamp: segment-level (the subtitle block) and word-level (each individual token). Word-level timestamps are stored in the caption JSON alongside each word, powering the click-to-seek interaction in the editor.
The word-level data model
Each subtitle block in CutSnap's caption format contains an array of word objects: { word: string, start: number, end: number, confidence: number }. The confidence score is exposed as a visual highlight in the editor — low-confidence words are underlined in amber, drawing your attention to likely transcription errors before you even listen to the audio.
Smart block splitting
When you split a subtitle block at a word boundary, CutSnap sets the new block's start time to the word's start timestamp and the previous block's end time to the preceding word's end timestamp. The gap between them is preserved, and neither block ever has overlapping timestamps. This keeps the subtitle SRT file valid and ensures the video player does not display two blocks simultaneously.
Does every word have a timestamp?
Yes, when the video is transcribed with word-level mode. Occasionally the model merges short function words into adjacent timestamps, but CutSnap interpolates in those rare cases.
What does the amber underline on a word mean?
Amber-underlined words have a confidence score below 0.8. They are likely to contain transcription errors and should be reviewed.
Can I adjust the start and end time of a single word?
Yes. Click on any word in the timeline view to select it and you will see numerical start/end inputs that you can edit directly.