Remove Background Noise and Reverb From a Voice in iZotope RX

Why cleanup comes before everything else

Before any EQ, compression, or delivery loudness target, I clean. The order is not negotiable: a compressor on a still-noisy voice pulls the hiss up between words, and an EQ that flatters the highs makes sibilance and clicks more present. You fix the problem at the source, you don't paper over it downstream.

Cleanup is subtraction. You remove what should not be there, noise, echo, clicks, so that only the voice remains. All the creative work, tone, presence, dynamics, comes afterward, on a clean foundation.

Each problem has its own module

RX is not one button: it is a toolbox where each problem has its own module. Mixing them up is the most common mistake. You don't reduce reverb with a denoiser, and you don't treat hiss with the click module.

• De-hum removes electrical mains hum, at 50 Hz in Europe and 60 Hz in North America, along with its harmonics.
• De-click and Mouth De-click remove digital clicks and mouth noises, saliva and tongue clacks respectively.
• De-plosive catches plosives, the popping p and b sounds left when there was no pop filter at the source.
• Voice De-noise, or Spectral De-noise which learns a profile from a silent passage, lowers hiss and constant background noise.
• De-reverb reduces reverb and room echo, the number-one problem with recordings made at home.
• De-ess tames sibilance, and Breath Control attenuates breaths without removing them completely.
• Spectral Repair, in the spectral editor, erases a one-off noise by hand, a cough, a beep, a creaking chair, by painting it out directly on the spectrogram.

The order that avoids artifacts

Order matters as much as settings. You go from the most localized and tonal to the most diffuse: anything one-off or at a fixed frequency comes out cleanly before you touch broadband noise, otherwise the denoiser works on a still-polluted signal and leaves traces.

My usual chain: De-hum, then De-click and Mouth De-click, then De-plosive, then Voice De-noise, then De-reverb, then De-ess, and finally Breath Control. Spectral Repair steps in on demand, for isolated accidents. Each stage is done by listening to the result, not by stacking modules blind.

Dialogue Isolate: when one module is enough

For a few versions now, RX has offered a shortcut: Dialogue Isolate. Instead of chaining denoise then de-reverb, this module separates the voice from noise and reverb in a single pass, using a neural network, offline or in real time. It is the tool most tutorials lead with, and it is stunning on a badly degraded take.

I use it as a starting point, not as a final setting. On an already decent voice, the module-by-module approach stays finer: you remove only what is actually a problem. Dialogue Isolate shines when there is nothing to lose; targeted work shines when the voice is already almost good. Note: Dialogue Isolate needs the Standard or Advanced edition of RX, not Elements.

Subtractive and light: the golden rule

The temptation is to push everything to the maximum. That is exactly what gives away an amateur cleanup: a voice denoised too hard turns metallic and underwater, with artifacts that shimmer in the highs, what we call warble. A few decibels of reduction you can't hear beat a perfect silence bought at the cost of a degraded voice.

I set each module to the minimum that works, then compare with and without, A/B bypass, at matched volume. If the bypass is barely audible but the problem is gone, it worked. If the voice changes color, I back off.

What RX does not do

RX repairs a lot, but it does not replace a good take. A clipped recording does not truly un-clip: the De-clip module reconstructs an approximation of the peaks, not the lost signal. A very reverberant room always leaves a trace after De-reverb. And no module rescues a badly placed mic or an input level so low it drowns in hiss.

The AI-assisted modules, Voice De-noise and De-reverb, give an excellent starting point, but the final call stays with the ear. I use them as a first pass, not a verdict. I don't claim miracles: on a genuinely damaged take, I say so and suggest re-recording the passage rather than delivering a voice stripped of its character.

One-click AI tool or iZotope RX: which to choose

A wave of online tools promises to remove background noise for free, in one click. For a story, a video call, or a social clip, they are more than enough, and I'll say it plainly: paying for pro software for that makes no sense.

RX earns its place when the stakes rise: an audiobook that has to pass the ACX technical check, a documentary dialogue, a branded voice-over. There, the one-click tool leaves artifacts (metallic voice, half-removed reverb, eaten sibilance) that an attentive ear catches at once. RX lets you dose problem by problem and stop before the voice loses its character. The real question is not free or paid, it is how far the result has to hold up.

Where it fits in the workflow

RX integrates with Pro Tools, either as a plug-in directly on the track, or via RX Connect for a round-trip to the standalone app when a deeper cleanup is needed. Cleanup ends up in the same session as the edit and the mix: everything stays in the same hands, with no intermediate export that degrades the file.

Once the voice is clean, and only then, comes the rest: EQ, compression, then setting the loudness to the delivery target, for example around -16 LUFS for a podcast, or the RMS values ACX requires for an audiobook. Clean first, calibrate second: never the other way around.