Paraphrase vs Audacity: cutting speech by words or on a waveform

Bottom line: they do different jobs. Audacity is a free, open source audio editor with noise reduction, compression and effects, and you cut by selecting regions of a waveform. Paraphrase transcribes your recording and you cut by clicking words, with an AI pass that proposes the cuts, but it does no audio processing at all. If your problem is the sound, use Audacity. If your problem is the hours spent finding ums and restarts, use Paraphrase for the cut and Audacity for the polish.

Audacity has been around a long time: its homepage counts "100M+ downloads since 2000". If you are here because editing speech in it is slow, it is worth being clear about why, because that is exactly where Paraphrase takes a different approach.

Quick comparison

What editing speech in Audacity actually is

Audacity edits sound. You record or import, and you see a waveform. To remove a section you select it "by clicking and dragging in the waveform" and press Delete. Audacity's own Cut page suggests using that command to "Remove ums, coughs, and long pauses from spoken recordings".

For speech, that means:

• Listening through the recording to find each filler, false start and tangent.
• Zooming in to place the selection edges cleanly.
• Cutting, checking, and moving on, with unlimited undo when a cut goes wrong.

Where waveform editing hits its ceiling for speech

A waveform shows loudness over time. It does not show words. An "um" and a short word can look much the same; a false start looks like the rest of the sentence; the same point made twice looks like two perfectly good paragraphs. To find them you have to listen, which for an hour long interview means at least an hour of listening before you have cut anything.

That is not a fault in Audacity. It is a general audio editor, built for music, restoration and everything else. Cutting speech by meaning is a narrower job.

What Paraphrase does differently

Paraphrase starts from the words:

• Every word has a timestamp. Your recording is transcribed, and the transcript is the editor. Click a word to cut it, click again to restore it, shift-click to cut a run, drag a word to move it. Cut words stay on screen, struck through.
• Clean up with AI proposes the cut. One button. A language model reads the transcript and marks filled pauses, false starts, hedges and repetition, including the same point made twice in different words. It is a proposal: any word can be put back. It is tested on English speech.
• The export is your recording, cut. Pieces are removed at word boundaries and joined back up. WAV and FLAC are lossless, so a cut of a lossless master stays lossless.
• It never generates speech. No voice cloning, no overdub, no text to speech. A word you never said cannot appear in the export, because there is no audio for it.
• Non-destructive and in the browser. The original is never changed, edits save as you go, and there is nothing to install.

You can read more in the editing docs and the AI cleanup docs.

What Paraphrase does not do that Audacity does

If your recording has fan hiss or mains hum, or needs compressing and levelling, Paraphrase will not help. Audacity will, for free. Paraphrase also has no multitrack, effects or plugins, and it is not free beyond the 15 minute allowance.

Using both

This is the setup we would actually suggest for spoken word that needs real cleanup:

1. Record or upload into Paraphrase, or paste a link. Premium projects hold up to 4 hours of recordings.
2. Cut by the words. Run Clean up with AI, review what it struck through, put back anything that matters, and make your own cuts and reorders.
3. Export WAV (lossless) and open it in Audacity.
4. Finish the sound in Audacity: Noise Reduction from a noise profile, then a Compressor preset and the Limiter.
5. Export the final file from Audacity in whatever format you publish.

Because Paraphrase is non-destructive and re-exporting costs nothing, you can go back and change the cut later and run it through Audacity again.

Why trust this comparison

Everything above about Audacity comes from Audacity's own pages, all read on 17 October 2026: the home page, the features pages for cutting and noise reduction, the FAQ, the download and Audacity 4 pages, and the support manual's pages on editing, saving and exporting, FFmpeg and the compressor. Paraphrase facts come from our own code and pricing, checked the same day.

We make Paraphrase, so this is not a neutral comparison.

Where Paraphrase leads, and where it does not. It never generates speech, so an export can only contain words you said. Every AI cut is shown in the transcript and can be restored word by word. Credits are spent only at transcription, so cleanup, re-editing and re-exporting are free. It is also the narrower tool: no remote recording, no noise removal, no captions, no AI voice and no collaboration, and unlike Audacity it is not free beyond its starting allowance.

We have no customer reviews, ratings or testimonials to show you, because we have not collected any. Judge it on one of your own recordings instead.

So which should you use?

Use Audacity on its own if your edits are light, your budget is zero, or most of your work is on the sound itself. It is mature, free and genuinely capable, and our Audacity review covers it in more depth.

Add Paraphrase when cutting speech is what eats your time: long interviews, lectures, courses or voiceovers where finding the ums and restarts takes longer than recording did. The podcast editing and interview editing pages show it on those jobs.

If neither fits, the Audacity alternatives roundup covers the other options, and Is Audacity free? covers what Audacity costs and where money can come in.

Paraphrase's free plan gives you 15 minutes of transcription, three AI cleanups a day and projects up to 5 minutes, with no card. Create a free account at paraphrase.bot and cut one of your own recordings by its words, then finish it in Audacity.