Audacity remains the free editing tool that millions of podcasters and musicians rely on. Yet if you’ve used its built-in noise reducer or vocal-isolation effect, you know they can sound surgical. True AI plugins behave differently. Many are trained on tens of thousands of audio examples, so they can separate a voice from an air-conditioner rumble with impressive accuracy. Thanks to Audacity’s VST3 support, installing these tools is easier than you might think. Here are the plugins I’d start with and the exact steps to load them.
How VST3 support opened the door for Audacity AI plugins
Before Audacity 3.1, effects could only talk to Audacity through older VST2 ports or Nyquist scripts. Most neural-network-driven audio tools never shipped in that format. Since Audacity 3.1, the software loads VST3 plugins on Windows, macOS and Linux. That means you can now use machine-learning-based tools that would never have run inside Audacity a few years ago.
These plugins behave like ordinary Audacity effects. You select part of a track, open the effect from the Effects menu, tweak its parameters and press Apply. Audacity renders the result into the waveform, which is a nice advantage: you don’t need a top-end PC because an AI effect only has to process as fast as the render, not keep up with live playback. The Preview button lets you audition a short section first, so you can adjust the settings on a manageable bite of audio before committing.
Installing VST3 plugins in Audacity: no special tricks
The process is the same for nearly every AI plugin, no matter where you download it.
- Install the plugin with its own installer. On Windows this places the .vst3 file in C:/Program Files/Common Files/VST3. On macOS, look under /Library/Audio/Plug-Ins/VST3.
- Launch Audacity and click Effects > Plugin Manager.
- Find your new plugin in the list. If it doesn’t have a green check mark, click Enable.
- Click OK and restart Audacity if the manager asks you to.
- Open the Effects menu again. The plugin appears near the bottom in the All Effects section.
If Audacity still doesn’t show the plugin, go to Preferences > Effects and make sure the VST3 path hasn’t been changed. Some paid plugins require an activation app before the plugin is unlocked; that’s normal. You’ll see the same folder, but the plugin will only load after you’ve logged in to the vendor’s license manager.
The Audacity AI plugins worth your time
Countless developers call their product ‘AI’ when it’s just an adaptive filter. The list below sticks to tools that are actually running a neural network and that work cleanly as VST3 effects inside Audacity.
Acon Digital Restoration Suite 2
Acon Digital has one of the more mature restoration kits for dialogue and field recordings. Restoration Suite 2 includes DeNoise 2, DeHum, DeClick and DeBuzz, and the engine learns from a short noise sample rather than applying a one-size-fits-all mask. If you’re cleaning up an old cassette recording or a voice track recorded next to a fridge, this suite is hard to beat. The spectrogram in DeNoise 2 makes it a natural fit for Audacity users who are used to seeing audio. It’s not free, but the trial lets you process a few seconds and buy a full licence if it solves your problem.
Waves Clarity Vx
Waves Clarity Vx is built for one job: cleaning speech. It uses a trained model that distinguishes a human voice from room tone, keyboard clicks, fans and even certain kinds of background chatter. In Audacity, you select a noisy voice clip, open Clarity Vx and raise the Reduction control gradually. Unlike Audacity’s built-in noise removal, Clarity Vx rarely introduces that hollow, underwater quality. It works best on mono 44.1 kHz or 48 kHz voice tracks. You’ll need Waves Central and a free Waves account to activate it. The plugin often goes on sale for around $29, so it’s worth adding to your wishlist if you record vocals or interviews.
Neural Amp Modeler (NAM)
If you record electric guitar, skip the usual amp sims and try Neural Amp Modeler instead. It’s a free, open-source VST3 plugin that plays back captures taken from real tube amps. Each capture file is owned by the neural network that modelled the amp’s exact response. You load it in the same way you’d load an impulse response, but the dynamic result is much closer to playing a real amplifier. You can feed it a clean DI guitar track recorded into Audacity and preview how it sounds through dozens of community-made captures. This is one of the clearest examples of an AI tool that isn’t just a copycat EQ.
sonible smart:EQ 4
sonible builds plug-ins around listening models, and smart:EQ 4 is their neural-network-based equaliser. It analyses a selected clip, places a suggested EQ curve over the frequency spectrum and then lets you adjust the strength. In Audacity, I find it most useful for de-mudding voice recordings before compression or limiting. It can also tuck tricky sibilance into place better than a static de-esser in many cases. smart:EQ 4 is a paid plugin with a trial, and it loads as a standard VST3 effect. Since Audacity applies effects offline, you can preview different strength settings without risk and choose what works for that particular recording.
When an offline AI service is the right alternative
You don’t always need a plugin for AI audio processing. Some of the most powerful tools are cloud services that accept an audio file and return a cleaned one. Adobe’s Podcast Enhance is a free web tool that removes room roar and balances speech surprisingly well. You export a WAV from Audacity, drag it onto the Enhance page and wait a minute for the file to process. Then you import the enhanced file back into Audacity with File > Import > Audio. LALAL.AI works similarly for stem separation, though the free tier limits how many minutes you can process.
The main difference is speed and privacy. A web service sends your audio to someone else’s server, which is fine for a demo but not ideal for unreleased commercial work. Local VST3 plugins keep everything on your machine. That privacy is one of the strongest arguments for the plugins above if you do client work or produce copyrighted music. Start with the trial of each tool, compare the results and let your ears and your timeline decide the winner.