What counts as a filler word?
In practice it is a small set of sounds and habits that carry no meaning: um, uh and ah, the stretched "so..." at the start of an answer, repeated false starts where a sentence restarts two or three times, and personal verbal tics such as "you know" or "like" used as punctuation.
Pauses are not filler. A speaker taking a breath before a point, or leaving a beat after one, is doing something useful. Removing silence along with the ums is how edits end up sounding rushed and airless.
Why removing every filler word sounds unnatural
Speech is not written prose read aloud. Listeners tolerate a surprising number of hesitations without noticing, because that is how conversation sounds. What they do notice is the absence of them: perfectly clipped sentences with no breath between, or an interview where every answer begins instantly. It reads as processed, and in an interview it can make a thoughtful guest sound glib.
The useful test is whether a filler is doing any work. An um that fills the moment while someone thinks is doing work. Three ums in one sentence is not.
Manual removal
The traditional approach is timeline editing: find each filler in the waveform or the transcript, cut it, then close the gap by the right amount rather than all of it. Leaving a fraction of the original pause is what keeps the edit invisible. Modern editors help by letting you delete words from a transcript, but the judgement is still yours, and on a long episode this is slow work.
Automatic filler-word removal
Automated removal is good at the mechanical part: finding the common fillers and taking them out consistently across a long file without you scrubbing through it. What it cannot do is understand your intent. It does not know that a particular hesitation was the moment the interview turned, or that a false start carried the emotion.
So the sensible workflow is automation first, then a listen. If a section feels clipped, that is the place to intervene, and it is usually a handful of moments rather than the whole file.
Podcasts, voiceovers and audiobooks expect different things
- Podcasts: conversational shows can carry plenty of natural hesitation. Trim the worst of it, especially clusters, and stop there.
- Voiceovers: commercial and corporate reads are usually expected to be clean, so tighter removal is appropriate.
- Audiobooks: be careful. Narration is a performance matched to a manuscript, and automatic editorial changes are not always welcome. Most narrators handle fillers by re-recording the line, and mastering stays purely technical. Our audiobook mastering page explains where that line sits.
Using Optivox
Filler-word removal in Optivox is optional and off unless you choose it, because whether it is appropriate depends entirely on the kind of recording. Turn it on for a voiceover cleanup or a tidy-up of a podcast episode, leave it off for narration where the read should stay exactly as performed. Everything else, the noise reduction, levelling and mastering, runs either way.
The how it works page shows where the option sits in the flow.