Tools and Recommendations

The best free noise reduction plugins for voice (and when stock REAPER is enough)

By StudioRescue Editorial

Published September 15, 2026 · Updated September 15, 2026

“Which free noise reduction plugin should I use?” is a question the search results answer badly. Most lists rank ten tools for every kind of audio at once, put paid plugins under a “free” headline, lean on affiliate links, and rarely say what any of the tools actually does under the hood. That last part is the part that decides whether a plugin helps your recording or wrecks it.

This is a shorter list, written for one job: a voice recording with unwanted background noise, in REAPER, with no budget. Everything below was checked on the vendor’s own page on September 15, 2026. Prices and licences change, so check before you download.

First, the three kinds of noise reducer

Almost every denoiser you will find is one of three designs, and the design tells you where it will fail.

• Spectral subtraction. The plugin splits the sound into frequency bands using an FFT, learns what the noise looks like across those bands, then subtracts that profile from everything. Very effective on steady noise. Push it hard and you get the classic metallic, watery, underwater artifacts, because the subtraction leaves ragged fragments behind. ReaFIR works this way.
• Broadband dynamic reduction. The plugin works in the time domain and pulls down whatever is below a threshold, band by band, without an FFT. Fewer smearing artifacts and no added latency, but it is blunter, and it can breathe or pump if the noise level moves. Bertom Denoiser and the one-knob voice tools work this way.
• A trained speech model. The plugin has a model of what human speech sounds like and rebuilds the voice while discarding everything else. It removes far more noise than the other two and it does not need a noise-only sample. The catch is that it only knows speech, so anything musical, breathy, or unusual in the performance can get thrown out along with the noise. RNNoise works this way.

Start with the one you already have

REAPER ships with ReaFIR, and Cockos describes it as an FFT-based dynamics processor with four modes: EQ, Compressor, Gate, and Subtract. Subtract mode is the one that matters here, and Cockos describes it plainly as being able to build noise profiles and subtract noise from the signal. FFT sizes run from 128 to 32768, which is a wider range than most dedicated denoisers give you.

In practice you play a section that contains only the noise, let ReaFIR build the profile, then switch the profile building off and play the whole take. That is the entire method, and it is genuinely good on hiss, fan noise, and air conditioning. The full step-by-step is in our guide to reducing background noise on a podcast recording in REAPER, and if the problem is specifically mains hum rather than broadband noise, removing hum is a different and easier fix.

Try ReaFIR first, on the actual recording, before you download anything. A surprising number of people install three plugins to solve a problem the stock one already solved.

How to judge a free denoiser in about two minutes

Free plugins are easy to collect and hard to compare. These four checks separate the useful ones fast.

Four free options worth having

ReaFIR (Cockos) – the one to try first

What it is: an FFT-based dynamics processor bundled with REAPER, with a Subtract mode built for noise profiles.

Good at: steady, unchanging noise. Hiss, computer fans, air conditioning, a room tone that does not move. The very large FFT size range gives you a real tradeoff to work with, since bigger sizes resolve low-frequency noise better while smaller sizes smear transients less.

Where it falls down: it needs a noise-only section to profile, which a tightly edited recording may not have. It adds latency, so it is a mixing tool rather than a monitoring tool. And it produces the classic subtraction artifacts if you raise the profile too far.

Cost: included with REAPER. Cockos also publishes a standalone ReaPlugs bundle for use in other hosts, which includes ReaFIR and ReaGate. Be aware that the standalone bundle is version 2.36 and dated January 2, 2016, and it is VST2 only. Inside REAPER you get the current build, so this only matters if you were planning to install ReaPlugs into a different DAW.

Bertom Denoiser Classic – zero latency, no FFT artifacts

What it is: a noise reduction plugin from Bertom Audio, currently version 3.0.11, offered pay-what-you-want including free. Bertom states it works with zero latency, low CPU use, and specifically without FFT and the artifacts that come with it. It is not limited to voice.

Good at: being usable in more places than ReaFIR. Zero latency means you can monitor through it while recording. It is also a better fit when the recording has no clean noise-only section to profile.

Where it falls down: a single global threshold is a blunt instrument when the noise is uneven across the frequency range, and it is less surgical than a good spectral tool on very steady noise. The paid Denoiser Pro at $29.99 adds per-band thresholds and an adaptive mode for noise that changes over time, which tells you exactly what the free version does not do.

Formats: VST3 and AAX on Windows 7 and later, VST3, AU, and AAX on macOS 10.13 and later including Apple Silicon, and VST3 on Linux. No VST2 build, which is fine for REAPER.

Viator Voice De-Noiser – one knob, no decisions

What it is: a free single-control noise reducer from Viator DSP, version 2.0.0, listed as free with no copy protection. Input and output trims sit either side of the one reduction control.

Good at: speed. Podcast, stream, and voiceover work where you want the fan noise down and you are not going to spend fifteen minutes on it. One control means you cannot set it badly wrong in a complicated way.

Where it falls down: one control also means you cannot fix a specific problem. If the reduction that quiets the noise also dulls the voice, there is nothing to adjust except how much of it you use. It is voice-tuned, so do not reach for it on instruments.

Formats: 64-bit VST3 on Windows, Audio Unit and VST3 on macOS as a universal build for both Intel and Apple Silicon.

RNNoise (Noise Suppression for Voice) – the heaviest free reduction

What it is: a free, open-source plugin built on Xiph’s RNNoise speech model, published by werman. It is distributed in VST2, VST3, LV2, LADSPA, AU, and AUv3 for Windows, macOS, and Linux. The most recent tagged release is v1.21, which adds an updated model and wet/dry controls, though it is published as a pre-release; v1.10 is the release currently flagged as latest.

Good at: removing a lot of noise with no setup and no noise profile. On a noisy spoken-word recording, a laptop mic, or screen-capture audio, it will do things the other three cannot.

Where it falls down: it was designed for real-time voice communication, not for music, and it behaves like it. It can flatten a sung vocal, strip breaths, and make a voice sound synthetic. The wet/dry control in the newer build exists precisely so you can blend the effect back rather than using it flat out, which is how most people should use it on anything they care about. Treat it as a strong tool for speech and a last resort for singing.

A quick way to pick

Two practical notes before you download

New plugins do not always appear in REAPER straight away. If a plugin is installed but missing from the FX browser, it is usually a scan or a folder path issue rather than a broken download, and the install and rescan steps are covered here.

Second, treat “free” as a fact with an expiry date. Pay-what-you-want plugins become paid, free tiers get folded into subscriptions, and developers stop maintaining things. Everything above was verified on the developer’s own page on the date at the top of this article, and not on a download-aggregator site. Do the same before you install.

What free noise reduction will not do

This is the part most lists skip, and it is the part that saves you an afternoon.

• It will not remove room echo. Reverb is your own voice arriving late, so it is not noise and there is nothing steady to subtract. Pushing a denoiser at a roomy recording makes it sound worse, not drier. That is a separate problem with a separate approach.
• It will not remove other voices, music, or traffic that changes. Anything that moves defeats a noise profile, and a speech model will happily keep the other person’s speech too.
• It will not fix mouth clicks or lip smacks. Those are short, sharp events sitting inside the voice, not a steady background, and denoisers walk straight past them. They need editing rather than processing.
• It will not undo clipping or distortion. Nothing here reconstructs a waveform that was destroyed on the way in.
• It will not make a noisy recording sound like a quiet one. Every denoiser removes some of the voice along with the noise. The goal is a recording where the noise no longer distracts, not a silent background.

If it still sounds wrong

1. You used too much. Halve the reduction. Most people set a denoiser by how much noise disappears, which is the wrong thing to listen to. Set it by how much the voice survives.
2. Your noise profile included speech. If the profiled section had even a breath in it, ReaFIR is now subtracting part of the voice. Reprofile on a genuinely silent gap.
3. It is low rumble, not broadband noise. If the problem is under the voice rather than behind it, a high-pass filter does more than any denoiser will, and does it without artifacts.
4. It is the wrong design for the noise. Moving, irregular noise will not respond to a subtraction tool no matter how you set it. Switch approaches rather than pushing harder.
5. You are stacking plugins. Two mild denoisers in series is not gentler than one. The artifacts compound. Pick one, set it conservatively, and stop.