Audacity 3.7.3 review: noise reduction and export, tested on Linux

By ToolBrivio Editorial · Hands-on test of Audacity 3.7.3 on Linux, October 14, 2026, plus official documentation where marked. We earn no affiliate commission from links on this page. Research approach · Link disclosure.

We tested Audacity 3.7.3 (Debian package 3.7.3+dfsg-1) on Debian 13, running under a headless X server. The job was narrow and common for interview editors: take a WAV with steady background noise, capture a noise profile, apply Noise Reduction, and export WAV and MP3. We measured the files before and after with sox and ffmpeg rather than judging by ear.

Review basis: Everything in the test sections below was run on Audacity 3.7.3. We did not test Audacity 4.x; the short 4.x section near the end is based only on official documentation. Our test file was synthetic (two tones with tremolo plus pink noise), not a recorded voice, so it tells you how the effect behaves on a steady noise bed, not how a particular speaker will sound.

Verdict

For steady hiss or room tone, Audacity 3.7.3's Noise Reduction does what its settings say: with a 12 dB setting, the noise-only section of our file dropped by 12.0 dB RMS. The catch we measured is that the wanted signal also lost 5.2 dB RMS, so treat the effect as a trade, not a free cleanup. Exports to WAV and MP3 worked from the scripting pipe without extra setup. If you need to automate the noise-profile step itself, expect to touch the GUI: our scripted NoiseReduction: call returned “OK” but we had to capture the profile through the dialog.

• Pro: Predictable floor reduction: −43.69 dB → −55.69 dB RMS on the noise-only section with the 12 dB setting.
• Pro: mod-script-pipe ships with the Debian 3.7.3 package; import, select and export all worked from a script once the module was enabled.
• Con: On our test signal, the tone section also dropped from −23.93 to −29.09 dB RMS. Listen to speech after processing and check its level.
• Con: The profile step (“Get Noise Profile”) needed the dialog in our run; the module is disabled by default and has to be switched on first.
• Limit: The default WAV export wrote 16-bit PCM from our 32-bit source. Choose the encoding yourself if you need to keep bit depth.

Test setup

We generated the input with sox: 2 seconds of pink noise only, followed by 6 seconds of 220 Hz + 330 Hz tones with 4 Hz tremolo, mixed with the same level of pink noise (44.1 kHz, mono, 8.000 s). Audacity ran with its own config directory so it would not touch another user's settings.

1. Enable scripting: in audacity.cfg, under [Module], set mod-script-pipe=1 (the first launch wrote it as 4, which left the pipe closed). After a restart, /tmp/audacity_script_pipe.to.1000 and .from.1000 appeared. The in-app route is Edit > Preferences > Modules.
2. Import over the pipe: Import2: Filename=…/input_noisy.wav, then Select: Start=0 End=2 Mode=Set to select the noise-only section.
3. Effect > Noise Removal and Repair > Noise Reduction… > Get Noise Profile.
4. SelectAll:, reopen Noise Reduction, set Noise reduction (dB) to 12, keep Sensitivity 6.00 and Frequency smoothing 6, choose Reduce, then OK.
5. Export over the pipe: Export2: Filename=…/output_nr.wav NumChannels=1 and the same with output_nr.mp3.

Results

WAV figures come from sox … stats. The sox build on our test box had no MP3 handler (“no handler for file extension `mp3'”), so the MP3 was measured with ffmpeg's astats filter. The MP3 is 45.7 ms longer and starts 25 ms late in ffprobe's view. That is ordinary encoder padding, but it matters if you line the MP3 up against video or subtitles.

What failed or needed a workaround

• Scripted Noise Reduction. Sending NoiseReduction: through the pipe with the noise section selected returned “BatchCommand finished: OK”, yet the screenshot afterwards showed noise still in place and Repeat Last Effect grayed out. In this run the profile step had to go through the dialog.
• Scripting is off by default. The pipe did not appear until we changed mod-script-pipe from 4 to 1 and restarted.
• The first launch shows a modal. “Welcome to Audacity!” / “What's new in Audacity 3.7.3” has to be dismissed before automation is reliable.

Who it suits

Audacity 3.7.3 fits an editor who has a recording with constant background noise and wants direct control over the amount of reduction. It is free desktop software under the GPL, according to the official FAQ. If you want hands-off finishing, compare our Auphonic review. For transcript-led cutting, look at Descript. Neither was part of this test.

Practical advice from the run: take the noise profile from a clean pause at least a second long, start at a moderate dB setting, and compare speech level before and after. On our file a 12 dB setting removed 12 dB of noise but also took 5 dB off the wanted signal.

Working with an existing caption track? See how to fix subtitles that drift after audio edits before you shorten the recording.

Audacity 4.x and .aup4 (not tested, documentation only)

We did not test Audacity 4.x. This section is based only on official documentation, checked October 6, 2026. According to the official changelog, 4.0 shipped on September 3, 2026, and 4.0.1 on September 30. Version 4 saves projects as .aup4 and can convert .aup3 projects, leaving the original unchanged, but cannot save back to .aup3. The 4.0 compatibility notes list missing features, including Macro Manager, the scripting pipe, Time Tracks, Note/MIDI tracks, Mixer, VAMP/LADSPA hosting and Play-at-speed; the 4.0.1 notes do not announce their return. If your workflow depends on mod-script-pipe, as ours did, check those notes before you upgrade, and keep a copy of your .aup3 projects.

Local privacy tip (documentation only)

This tip is based on official documentation, not on our hands-on test. Audacity can keep a project on your computer without cloud saving, but that does not mean every function is network-free. The desktop privacy notice covers optional update checks, error reporting, an installation identifier and separate Audio.com services. Review those settings before connecting a project or sending a report. We did not test privacy behavior in this review.

Evidence

Test logs, screenshots, exported files, sox/ffprobe output and SHA-256 hashes are kept in ToolBrivio's evidence archive for this review (audacity, October 14, 2026).