The best tools to remove filler words from audio
Bottom line: if you want to see and approve every cut, and catch more than "um" (false starts, hedges, and the same point made twice), Paraphrase (ours) shows each proposed cut struck through in the transcript and never generates speech to smooth a join. Riverside and Descript both give you a reviewable filler list inside a full editor, Riverside with contextual fillers like "like" and "you know" and a "Remove all" button. In Adobe Podcast you find fillers in the transcript and delete them yourself. Audacity is the free, fully manual route.
Filler words are the sounds and phrases people use while thinking: "um", "uh", "hmm", and in some contexts "like", "you know" or "I mean". Removing all of them is rarely right. One of Adobe's editing guides puts it well: "Removing every filler word might sound unnatural, though, so make sure you pick and choose which ones to cut." So the useful questions are how a tool finds them, what else it catches, and how easily you can check its work.
Filler word removal compared
| Tool | How it finds fillers | Catches beyond um and uh | How you review | Generates speech | Free to try | Entry paid price |
|---|---|---|---|---|---|---|
| Paraphrase | Language model reads the transcript in context, no word list | False starts, hedges, repetition, restated points | Every cut struck through, click to restore | Never | 15 minutes, 3 cleanups a day | £13.99 a month |
| Riverside | Detected and highlighted in the transcript | "Like", "you know" only when used as fillers; breaths | Accept or discard each card, restore any | Yes, paid plans | Free plan | £24 a month yearly (£288), £35 monthly |
| Descript | Underlined in the transcript, listed in a sidebar | Retakes and Edit for Clarity are separate tools | Each instance with audio preview | Yes, AI speech and voice clones | Free plan | $16 a person a month yearly |
| Adobe Podcast | You, reading the transcript, which includes filler words | Whatever you choose to delete | You choose each deletion | Firefly Pro, the plan that now includes Podcast Premium, includes Generate Speech | Free plan | Firefly Pro $19.99 a month |
| Audacity | You, by ear | Whatever you select | Your own ears, unlimited undo | n/a | Free | No paid edition |
Prices as shown on each vendor's pricing page on 17 October 2026, in the currency shown, not converted. Riverside's are as shown to a UK visitor. Adobe's UK Firefly page shows Firefly Pro at £19.99 a month including VAT.
How this list is ordered: by how easily you can see, check and reverse every individual removal before you export, and then by how much it catches beyond plain "um" and "uh".
1. Paraphrase: every proposed cut visible in the transcript
Paraphrase is ours, so weigh this accordingly. Its cleanup is one button: a language model reads the transcript and proposes a cut for filled pauses, false starts, hedges and repetition, including the same point made twice in different words, and it can move words where a sentence reads better. There is no word list. It judges each word in context and keeps anything that carries meaning.
The proposal is the point. Every cut shows struck through in the transcript, and clicking any word puts it back, so you are approving an edit rather than trusting one. The result is always your original recording with pieces removed at word boundaries. Nothing is regenerated to smooth a join, because Paraphrase never generates speech. Cleanup, re-editing and export use no credits once a recording is transcribed.
Limits worth knowing: cleanup has been tested on English speech; the free plan allows three cleanups a day and Premium thirty; and there is no mouth-sound, breath or noise removal. More in how AI cleanup works.
Best for: spoken recordings where you want to approve each cut and catch more than ums.
2. Riverside: context-aware fillers inside a recording studio
Riverside's help centre says it "detects filler words and highlights them in the transcript, so you can listen to each one in context and decide whether to remove it or keep it". It separates plain fillers ("um", "uh", "hmm") from contextual ones ("like", "you know", "I mean"), which are "flagged only when used as a filler", and flags "harsh cuts" that may clip nearby words. You can accept or discard each, or remove or restore all, and its editor can mute words rather than cut them. On paid plans, Find Fluff suggests longer segments to remove, like redundant takes.
Best for: shows already recorded in Riverside. See our Riverside review.
3. Descript: a filler sidebar with per-word choices
Descript underlines filler words in the transcript and lists each instance, with timestamp and audio preview, in a sidebar. For each you can delete it, delete it and leave a gap, or "Ignore" it, which strikes it through in the script but still removes the audio. An "Avoid harsh cuts" setting skips fillers it cannot remove cleanly. Separate tools handle retakes and broader tightening (Edit for Clarity), and that one can optionally regenerate audio with AI speech to smooth edits.
The limits: filler detection works "in English transcripts only", and each run uses AI credits, up to 30 per file each time you apply it.
Best for: English recordings you are editing in Descript anyway. See our Descript review.
4. Adobe Podcast: fillers in the transcript, removed by you
Adobe Podcast Studio (labelled beta) transcribes "every word", and Adobe says its transcription "delivers more accurate results, including filler words". Its guides say the text-based editor "can help you easily identify filler words in your recording", and advise you to "pick and choose which ones to cut". You remove one by deleting it from the transcript.
Adobe does have filler detection in a different product: Premiere Pro, a separate desktop application, lets you filter a transcript for filler words or pauses and "Delete all". It is not part of Adobe Podcast.
Where Adobe Podcast is strong is Enhance Speech, which removes noise and echo, something Paraphrase does not do at all. See our Adobe Podcast review and Paraphrase vs Adobe Podcast.
Best for: recordings you are already cleaning up with Enhance Speech, where you are happy to choose each filler yourself.
5. Audacity: free and entirely manual
In Audacity you listen, select the "um" on the waveform and delete it; its own Cut page lists "Remove ums, coughs, and long pauses" as a use. It takes time, but every cut is yours, and it costs nothing.
Best for: short recordings and anyone with time but no budget. See our Audacity review.
Why trust this list
Each competitor detail comes from the vendor's own pages, read on 17 October 2026: Riverside's filler and Find Fluff help articles, Descript's Remove filler words and Edit for Clarity articles, Adobe Podcast's guides, plan comparison and FAQ (and Adobe's Premiere Pro help article on filler detection), and Audacity's Cut feature page.
Paraphrase is ours. It leads on reviewing each AI cut in the transcript and on never generating speech; it does not lead on mouth sounds, breaths or noise. We have no customer reviews to cite.
How to choose a filler word remover
- Lots of fillers, little time: Riverside's "Remove all", then restore the ones you want back.
- Not English: Descript's detection is English only, and Paraphrase's cleanup is tested on English.
- You want to approve every cut: Paraphrase, Riverside or Descript, or Adobe Podcast if you would rather find each one yourself.
- Rambling, not just ums: Paraphrase's cleanup, Descript's Edit for Clarity, or Riverside's Find Fluff.
- Video that must stay in sync: mute rather than cut, which Riverside offers.
Whichever you pick, listen back before publishing. To try word-by-word cleanup on your own voice, create a free Paraphrase account: 15 minutes of transcription, three cleanups a day, no card.