Writing

Filler Words: Why We Say Um and How to Keep Them Out of Text

Nemo · · 5 min read

Filler words are the small sounds and phrases we use while thinking out loud: “um”, “uh”, “like”, “you know”, “so”, “I mean”, “basically”. They’re a normal part of speech and help listeners follow us. In writing they’re noise. If you dictate, the fix isn’t to speak perfectly, it’s to remove filler automatically before the text reaches the page.

This guide covers what filler words are, why everyone uses them, and how to keep them out of your writing without changing how you talk.

What counts as a filler word?

Linguists split them into a few groups. In dictation, all of them end up in a raw transcript.

Type Examples What it does in speech
Hesitation sounds um, uh, er, hmm Holds your turn while you think
Discourse markers so, well, okay, right Signals a new point
Hedges like, kind of, sort of, basically Softens a claim
Interaction phrases you know, I mean, you see Checks the listener is with you
Repetitions “I think, I think” Buys time to plan the rest
False starts “We should, actually, let’s…” Corrects course mid-sentence

Not every “like” or “so” is filler. “I like this” and “so that we can ship” carry meaning. That’s why simple find-and-delete doesn’t work, and why context matters.

Why do we use filler words at all?

Because speaking is planning in real time. Your brain is choosing the next words while your mouth is still finishing the last ones. Filler fills those gaps and tells the listener you’re not done yet. Research on conversation shows people use them across languages, ages and levels of education. Every language has its own: “eh” and “o sea” in Spanish, “tipo” and “né” in Portuguese, “euh” in French.

So the goal isn’t to train yourself out of them. Public speaking coaches do help people use fewer on stage, but when you’re dictating a draft, fighting your own speech habits just slows you down.

Why are filler words a problem in dictation?

Speech recognition is designed to write down what you said. A good transcription engine, like Whisper, will faithfully capture:

“So, um, I think we should, like, we should probably push the launch to, uh, next Tuesday, you know, because the, the docs aren’t done.”

That’s accurate. It’s also not something you’d send. Editing it by hand takes long enough that many people give up on dictation for anything longer than a sentence.

How do you remove filler words automatically?

You need a step that understands context, which today means a language model. The model reads the transcript and rewrites it as the sentence you meant, without adding anything new.

The example above becomes:

“I think we should probably push the launch to next Tuesday, because the docs aren’t done.”

Notice what stays: “I think” and “probably” are hedges that carry meaning, so they remain. Only the noise is removed.

Lyre is a macOS menu bar app that turns your speech into clean, ready-to-paste text in the MacBook notch, removing filler words, false starts and repetitions with a local model while keeping your meaning and tone.

Everything runs on your Mac: Whisper transcribes the audio, and Qwen3 (through a bundled llama.cpp server) does the cleanup. Nothing is sent to a server. Our article on AI dictation explains what that cleanup layer should and shouldn’t change.

What’s the difference between removing filler and rewriting?

This is the line that matters:

Good cleanup keeps Good cleanup removes
Your vocabulary and phrasing um, uh, er
Hedges that carry meaning (“probably”) Filler “like”, “you know”, “I mean”
Your tone, casual or formal Repeated words
Every fact, number and name you said Abandoned starts of sentences

If a tool turns your casual note into corporate prose, or adds details you didn’t say, it’s rewriting, not cleaning. Always check the first few results from any tool against what you actually said.

Can you keep filler words when you need them?

Sometimes you do want exact words: quoting someone, transcribing an interview, keeping a verbatim record. In Lyre you can turn the polish step off to get the raw Whisper transcript instead.

How do you dictate with less editing afterwards?

  • Don’t self-correct out loud too much. If you restart a sentence, just keep going; the cleanup handles false starts.
  • Pause between ideas. It helps the transcript and the cleanup.
  • Say names and numbers clearly. Those are the details worth getting right.
  • Dictate one message or paragraph at a time.

If you’re new to dictating on a Mac, start with how to dictate on Mac.

What are the limits?

Lyre pastes rather than types: the clean text goes to your clipboard and you press Cmd+V. There’s no global shortcut yet and no live streaming of words while you talk. It needs an Apple Silicon Mac with macOS 13 or later, and costs $5 a month or $39 a year (7-day refund, no free trial).

FAQ

What are the most common filler words in English?

“Um”, “uh”, “like”, “you know”, “so”, “I mean”, “actually”, “basically” and “right” are among the most common.

Are filler words bad?

Not in speech. They’re a natural part of how people talk. They only become a problem when speech is turned into writing without editing.

Does removing filler words change what I said?

It shouldn’t. A good cleanup removes noise and keeps meaning, tone and facts. Read the result before you send it.

Can Lyre remove filler words in other languages?

The cleanup works on dictation in other languages too, and can also translate to English as you dictate.

Is there a way to remove filler words from a recording I already have?

Lyre is built for live dictation, not for processing existing audio files.

Download Lyre and dictate the way you normally talk.

Contact

Licence keys, billing, bugs and ideas. We reply by email.