Stop Cutting by Ear: See the Exact Words at Every Cut Point
Hover the razor in VoxStudio and see the subtitle split into "before | after" at the exact cut point. Word-level timestamps explained for people who edit talking-head video.
By David, Product Specialist at VoxStudio · · 6 min read
Key takeaways
- The razor tooltip shows the words before and after your cut, so you stop guessing by ear.
- It is built on word-level timestamps from speech recognition, accurate to a word but not perfect.
- One click cuts picture, audio and subtitle at the same point.
The problem: speech is easy to hear and hard to cut
Editing an interview, a lecture or a podcast video is mostly deciding where one sentence ends and the next begins. Without help, that means playing the same two seconds again and again. You cut a little early and clip a consonant. You cut a little late and an "um" survives. You zoom in, zoom out, and play it once more to be sure.
Waveforms help less than you would hope. In speech, words run together, and a quiet breath can look like a gap while a pause inside a word can look like speech. The waveform shows loudness, not language.
Where the time goes
- Re-listening — Five or six passes over the same phrase to locate one word boundary, repeated for every cut.
- Cutting mid-word — A cut a few frames off leaves a clipped syllable or a stray "th", which you only notice on playback.
- Subtitles out of sync — After you cut the picture, the subtitle still holds the whole sentence, so you split and retype it separately.
- Losing the thread — On a long timeline you stop reading content and start reading shapes, which is how wrong lines get cut.
The workflow: hover, read, click
Start from a clip that already has subtitles, for example a reviewed session sent to the editor, or captions you generated on your Mac. Choose the razor (scissors) tool in the timeline toolbar and move the pointer along the timeline.
As you move, the playhead follows the pointer, so the preview and audio scrub with you. Beside the pointer, a tooltip shows the subtitle at that spot split in two: the words before the cut, a green marker, and the words after it. Slide the pointer until the marker sits exactly between the two words you want to separate, then click.
The click cuts the video, the audio and the subtitle at the same frame, and the subtitle text is split to match the tooltip you just read. What you saw is what you get, with no second pass to fix the captions.
If the pointer is in the silence between two subtitles, the tooltip shows the line before the gap and the line after it. That is useful for trimming dead air: you can see what you keep on each side without playing it.
Three ways to find a cut point
| By ear | By waveform | Subtitle cut hint | |
|---|---|---|---|
| Replays per cut | Many | Several | Often none, then verify once |
| Risk of cutting mid-word | High | Medium | Lower; the marker sits between words |
| Subtitle stays in sync | No, edit separately | No, edit separately | Yes, split with the cut |
| Works on noisy audio | Hard | Hard | Depends on transcription quality |
Where it pays off most
- Removing a stumble — Cut before the false start and after the restart without replaying either.
- Pulling a quote — Find the first and last word of the sentence you want, cut at both, and keep the rest for later.
- Tightening pauses — Hover inside a gap and read both sides to decide how much to remove.
- Matching a script — Cut exactly where the written line ends so picture and text stay aligned.
How it knows where each word is
Speech recognition turns audio into text. That alone does not say when each word was spoken. For that you need word-level timestamps: a start and an end time for every word. There are two common ways to get them. Some recognition models predict timing as they transcribe. For others, a second model called a forced aligner takes the finished text and the audio and works out where each word sits.
VoxStudio stores those timings on each subtitle clip. When you hover with the razor, the editor converts the pointer position into a time offset inside the subtitle and sorts the words: any word that has finished by that moment goes in the "before" column, the rest go in "after". The green marker is the boundary between the two groups.
If a subtitle has no word timings, for instance because its text was rewritten, the editor falls back to an estimate that spreads the characters evenly across the clip duration. The tooltip still helps, but treat it as a guide and confirm by ear.
How accurate is it?
Accurate to the word in practice, but never guaranteed. Timings come from a model, so quality depends on the model, the language, the recording and the speaker. Clean speech is usually placed very tightly. Crosstalk, music under speech, heavy accents, fast delivery and noise can move a boundary or merge two words. For languages written without spaces, such as Chinese or Japanese, the unit may be a character or short phrase rather than a word, depending on the model.
So use the hint to get to the right place fast, then play the cut once. One verification pass is far cheaper than hunting for the boundary by ear.
FAQ
What are word-level timestamps?
They are start and end times for every word in a transcript. They let an editor know exactly when each word is spoken, which makes it possible to show or cut at word boundaries.
How do I cut a video at the exact word in VoxStudio?
Choose the razor tool, move the pointer over the subtitle until the tooltip shows the words you want on each side of the green marker, then click. The video, audio and subtitle are split at that point.
Are word timestamps 100% accurate?
No. They are predicted by speech models and depend on audio quality, language and speaker. They are usually close enough to find a cut quickly, but you should play the result once to confirm.
Do I need subtitles on the timeline?
Yes. The cut hint reads the subtitle clips on the timeline. Bring in a reviewed session or generate captions on your Mac first.
What happens if the subtitle text was edited?
Without stored word timings, the editor estimates the split position evenly across the subtitle, so the hint is less precise. Verify by ear.
Explore the Mac workflows
- Offline Transcription for Mac
Transcribe audio and video on your Mac with local speech models, speaker labels and timed captions. Built for Apple silicon and macOS 15 or later.
Cut where the words end.
Load subtitles into the VoxStudio editor, pick the razor and let the tooltip show you the cut before you make it. Download VoxStudio for Mac
David — David works on the VoxStudio product team and spends most days inside the video editor, turning real customer workflows into faster ones.