How to Remove “You Know” and “Like” From Audio

Um and uh are easy targets — they stand out, they pause the sentence, and cutting them is nearly always safe. Phrase fillers like “you know”, “like”, “sort of” and “I mean” are a different problem. They hide inside real sentences, they borrow the rhythm of normal speech, and “like” is sometimes a verb. That is why bulk one-click removal works for um and uh but needs human judgment for these. This guide covers a workflow that handles them properly: automatic detection plus a quick review pass so nothing that belongs there gets cut.

Why phrase fillers are harder to remove

Three reasons. First, they are context-dependent: “you know” can be a filler (“it was, you know, amazing”) or a genuine question (“you know where it is?”). Second, “like” is a real word — it works as a verb, noun, adjective and quotative (“he was like, no way”). Any automatic system that deletes every “like” will mangle your sentences. Third, removing them changes cadence: ums sit in natural pause points, but phrase fillers sit mid-sentence, so cutting one joins two fragments that may need a beat between them. Plan a preview pass. If you want to fix the habit itself rather than clean recordings, there is also a guide on saying fewer fillers next time you record.

The workflow: detect automatically, review like a human

1. Get the recording into one file. Export MP3 or WAV from your recorder or editor; for video, the MP4 works too — the tool transcribes the audio track. Anything up to about 30 minutes goes through in one pass.
2. Drop it into the tool. Open the filler word remover and drop the file in. Everything runs in your browser — nothing is uploaded, which matters for client calls and unreleased material.
3. Transcribe. An on-device AI model (downloaded once, around 40 MB, cached after) timestamps every word. Skim the transcript for misheard words before touching anything, especially jargon, names and accents.
4. Remove the obvious ones in bulk. Um, uh and clusters are safe to remove all at once. For phrase fillers, do it in a second pass: let the tool highlight every “you know”, “like”, “sort of” and “I mean”, then click through and keep the ones that carry meaning.
5. Preview every kept cut. Play each edited sentence. A removed “you know” can leave two clauses crashing together; if it sounds rushed, restore it or trim the surrounding pause instead. UmBreak has a dedicated guide for removing awkward pauses when the sentence needs air instead.
6. Export and finish elsewhere. Download the cleaned WAV and drop it into your editor or publish it directly.

Which phrase fillers to keep

The rule is simple: cut the tic, keep the tool. Cut “like” when it is a verbal shrug (“it was like, totally fine”). Keep it when removing it breaks grammar or meaning: verbs (“I like this”), quotatives in storytelling (“she was like, wait what”), and approximations (“like twenty people”). Cut “you know” at the start of sentences and in clusters (“you know, you know”). Keep it when it checks for agreement in an interview or conversation — that is conversational work, not filler. Keep “I mean” when it introduces a genuine correction; cut it when it just delays the sentence. A cleaned recording should still sound like you on a good day, not like a script you are reading.

Honest limits

No tool can reliably tell a filler “like” from a grammatical “like” by word lists alone — that decision needs your ears, which is why the review pass exists. Heavy accents and fast overlapping speech also raise the transcription error rate, so re-check the highlights in those parts. And removing fillers does not fix delivery: if the pacing drags, trim dead air first and read the full podcast cleanup guide for the right order of operations — noise, fillers, pauses, then levels.