How to cut your verbal tics without cutting the word where it means something

Last updated 2026-09-17 · 8 min read

What is the short version?

Every filler word is also a real word doing a real job somewhere else in the recording. Strip every single "so" out and the sentences that needed one to connect their two halves now begin abruptly, which sounds considerably worse than the tic ever did.

1. Tell Vizard Agent which word and where it is a tic.
2. Ask it to keep the instances that carry meaning.
3. Ask it to check the joins afterwards.

What do you need before you start?

The recording and an honest list of your own habits. Vizard Agent removes exactly what you name and nothing else, so listening back once and writing down the two or three words you genuinely overuse is the whole of the preparation.

• The recording. Unedited.
• The words. "So", "like", "basically", "right".
• Where they are tics. At the start, usually.
• Anything else to cut. A phrase you regret saying.
• How far to go. Conservative or thorough.

What do you type into Vizard Agent?

Describe the position to Vizard Agent, not just the word itself. A rule that catches every single instance is the easiest one to state and it produces a recording full of clipped-sounding sentences that is worse than what you started with.

Prompt

Variants worth knowing:

• "Leave the ones doing real work." The distinction.
• "Cut the word and the laugh." A specific, one-off removal.
• "Rather hear a tic than a bad join." Sets the tolerance.

What does Vizard Agent actually do?

Here is the Vizard Agent sequence on a real recorded talk cleaned up at the request of its own speaker. Notice that the filler removal happens late, only after the obvious false starts are dealt with, and that it is deliberately narrower than the instruction would have allowed.

1. Lists only the clear false starts and clipped fragments.
2. Checks exact word timings for each one before cutting.
3. Builds a conservative edit and a retimed transcript.
4. Renders the speech-led cut with the audio and picture together.
5. Reviews the export for abrupt cuts and timing problems.
6. Maps a reported jump cut back to the source speech around it.
7. Inspects the source region to decide whether to soften or restore.
8. Identifies the exact words to remove in the audio and the subtitles.
9. Separates the tic instances of the word from the meaningful ones.
10. Removes only the sentence-initial ones and re-renders.

Step nine is the whole job. The same three letters are a habit in one place and a conjunction in another, and the only way to tell is by where they sit in the sentence and what comes after them.

Step one sets the tone for everything after it. Listing only the clear cases means the edit starts from an obviously-safe position, and anything more aggressive is a decision you make afterwards rather than a default.

Step six is the cost of surgical cutting. Every removal creates a join, and a join in the middle of a sentence can show as a jump in the picture even when the audio is seamless.

What does the result look like?

You, sounding like yourself on a good day. The sentences still connect, the habit is gone, and the corrections you asked for specifically — a word, a laugh, a name spelling — are gone too, with the captions rebuilt to match.

When does this not work well?

Some tics turn out to be load-bearing. A speaker whose whole rhythm is built on "you know" sounds oddly clipped once Vizard Agent takes them out, and a recording where the fillers were covering thinking pauses will simply develop gaps where the thinking used to be.

• Tics that carry rhythm. Removing them flattens the delivery.
• Fillers over pauses. You are left with silence.
• Very frequent tics. Hundreds of joins in one recording.
• On-camera speakers. Every cut risks a visible jump.
• Conversations. Over-editing an exchange makes it sound rehearsed.

How do you fix a result that came back wrong?

Say whether what you can hear is the tic or the join. Those are the only two directions this can fail in, and Vizard Agent simply moves the threshold one way or the other rather than re-doing the whole edit from scratch.

• "It sounds chopped." Fewer removals, or the joins are softened.
• "You left a lot of them in." The rule widens beyond sentence-initial.
• "That one meant something." That instance is restored.
• "The picture jumps." A cutaway or a hold covers the join.

How does Vizard Agent compare to doing it yourself?

By hand you search the transcript for the word and delete every hit, because doing it case by case across an hour is not realistic. Then you listen back and restore about a third of them, which takes longer than the original pass.

Common questions

Which words should I name? The ones you hear yourself say. "So", "like", "basically", "right", "you know".

Will it catch every one? Only the ones matching your rule. Vizard Agent deliberately leaves every other instance alone.

How does it tell a tic from a real word? Vizard Agent goes by where it sits in the sentence and what follows it, not by the word.

Can I remove a specific thing I said? Yes. Name the moment and the words and Vizard Agent cuts just those.

Will the picture jump? It can. Vizard Agent checks the joins and can cover them.

Do the captions update? Yes. They are rebuilt from the retimed transcript.

Can it be more aggressive? Yes, if you ask. Conservative is the safer starting point.

What about breaths? A separate decision. Ask Vizard Agent; removing all of them sounds unnatural.

Does this work on a conversation? Carefully. Over-editing an exchange makes it sound scripted.

Can it just tell me how often I say it? Yes. Vizard Agent will count them, and the number alone often changes how you record.

What if a removal leaves a gap? Vizard Agent closes it, or holds the picture if closing sounds wrong.

Will my voice sound processed? No. Vizard Agent removes material rather than regenerating any of it.

Can I hear both versions? Yes. Ask Vizard Agent for the conservative and the thorough cut.

What if I laugh mid-sentence? Say so. That is a one-off removal rather than a rule.