Audacity review: what it is good at and where it runs out for speech

Bottom line: Audacity is a free, open source desktop audio editor for Windows, macOS and Linux, with real noise reduction, compression and effects. Where it runs out is cutting long spoken recordings: you do it by hand on the waveform. Keep it for the polish, and add a transcript editor if cutting speech is what takes your time.

We make Paraphrase, a transcript editor, so treat this as an interested party's view. Every claim about Audacity below comes from Audacity's own website and support manual, read on 17 October 2026. We did not test Audacity hands-on for this review, and we have no customer reviews to report.

What Audacity is

Audacity describes itself as "the world's most popular free software for recording and editing audio". It is open source under the GNU General Public License (GPL v3), runs on Windows, macOS and Linux (with x86_64 and ARM64 builds), and is desktop only: the FAQ says "There is no mobile version of Audacity available." The current release is 4.0.1, which rebuilt the interface on Qt. According to the Audacity 4 page, every project saved in Audacity 3 opens in Audacity 4.

What Audacity does well

It is free, properly. "Audacity has always been and will always remain free for everyone." No paid edition of the editor, no subscription. The optional costs are cloud storage on audio.com and third-party plugins, covered in Is Audacity free?.

The noise tools are real and well documented. Noise Reduction works from a noise profile: you select a stretch of background noise, capture it, then apply the reduction to the rest. Audacity is honest about its limits. It "works best" on steady noise like hiss, fan noise and mains hum; for "traffic, crowd chatter, or background music" other tools usually do better; heavy settings can sound watery; and it "cannot fully substitute for good recording conditions". There is also a Noise Gate, which silences audio below a volume threshold.

It has the finishing tools. A Compressor (destructive or realtime, with factory presets), and a Limiter, Graphic EQ and Filter Curve redesigned for Audacity 4. Plus reverb, echo, delay, distortion, pitch, speed and a vocal remover, and third-party VST3, LV2 and Audio Units plugins.

Audacity 4 changed clip editing. Clips can be selected directly and several at once with Shift+Click, trimmed and time stretched with handles, and split with a dedicated tool on the S key. Trimming is non-destructive. Sync-lock has been replaced by ripple editing, so "Delete and close gap (all tracks)" keeps a multitrack project in sync.

Exports are flexible. MP3, WAV, FLAC, Opus, M4A (AAC) and Wavpack among others, with FFmpeg as a separate install for M4A and WMA because of patent restrictions. Projects save as .aup3, or to audio.com with backups and versioning.

Where Audacity runs out for spoken word

These are not flaws in a general audio editor. They are what happens when you use one for the specific job of cutting speech.

You edit sound, not words. To remove a section you select it "by clicking and dragging in the waveform" and delete it. Audacity's Cut page suggests using Cut to "Remove ums, coughs, and long pauses from spoken recordings", and that is a manual hunt: listen, find, zoom, select, cut, repeat. Unlimited undo helps you recover from a bad cut; it does not help you find the next one.

Deciding what to cut is up to you. Audacity's noise tools clean the sound. Choosing which words go (false starts, hedges, the same point made twice) is an editorial job, and Audacity's Cut page makes clear that in Audacity you are the editor.

Desktop only. If you edit on a machine where you cannot install software, Audacity is not an option.

Where Paraphrase fills the gaps, and where it does not

Paraphrase is built for exactly the narrow job above. It transcribes your recording with a timestamp on every word, and every edit is a choice of which words stay:

Click a word to cut it, click again to restore it. Shift-click to cut a run. Drag a word to reorder.
Clean up with AI proposes the cut: filled pauses, false starts, hedges and repetition, including the same point made twice in different words. Cuts show struck through and any word can be put back. Tested on English speech.
Non-destructive: the original recording is never changed. Exports are WAV, FLAC, MP3, M4A and Opus, or MP4 when the project has a video take.
It never generates speech, so what you export is only what you said.
It runs in the browser, so there is nothing to install.

Where it does not fill the gap, plainly:

No noise removal, speech enhancement, levelling, EQ, music or effects. Everything in the "does well" section above is something Paraphrase does not do.
No multitrack and no separate per-speaker tracks, and no speaker labels.
No subtitles or captions.
Transcription covers 25 European languages, detected automatically.
It is paid beyond a small free allowance: 15 minutes of transcription free, then £13.99 a month or about £4.99 a month on the three year cycle (checked 17 October 2026).
It needs a connection. It is a web app, and transcription runs on Paraphrase's servers.

Who Audacity is right for

Audacity is right for anyone who wants a capable audio editor at no cost: recording, multitrack work, noise reduction, compression and effects, on Windows, macOS or Linux. For spoken word with light edits, it is all you need.

It is slow on its own if you regularly cut long recordings of speech, because the cutting is done by hand on a waveform. The common answer is to use both: cut by the words in Paraphrase, export a lossless WAV or FLAC, and finish the sound in Audacity. See Paraphrase vs Audacity for that workflow and Audacity alternatives for the other options. Our podcast editing page shows the Paraphrase side.

To try word-level cutting on one of your own recordings, create a free account at paraphrase.bot. No card needed.