How to Sound Good (Part 3) - The Art of Noise Reduction

This is part 3 of my “How to Sound Good” series. In Part 1 , we bought microphones. In Part 2 , we set up gates to mute the silence. Now, we have to deal with the actual noise that happens while you are talking.

If you work from a sound-treated studio with silent air conditioning and no neighbors, congratulations, I hate you. You can skip this post.

But if you are like me—working from Airbnbs with loud fridges, motorhomes with aggressive A/C units, or coffee shops with that one guy laughing too loud—this is the most important part of the series. Noise reduction is the difference between sounding like you are “on a call” and sounding like you are “in the room.”

Know your enemy

Before you start twisting knobs and buying expensive plugins, you need to know what you are fighting. In the world of the “Roam Office,” noise generally falls into three buckets.

1. Static noise (The “Hiss”)

This is consistent, unchanging sound. It’s the air conditioner that runs for 20 minutes, the hum of your computer fans, the buzz of a refrigerator, or electrical ground loop hum.

Computers represent sound as waves. Static noise looks like a predictable, repeating wave pattern. Because it never changes, it is very easy for software to identify it and kill it without hurting your voice.

2. Dynamic noise (The “Chaos”)

This is the unpredictable stuff. Dogs barking, babies crying, sirens passing by, keyboard clacking, or dishes clanking.

Historically, this was almost impossible to fix. Traditional software couldn’t tell the difference between a dog barking and a human shouting. Fortunately, we are living in the future, and AI has changed the game here.

3. Reflective noise (The Room Itself)

This is the trickiest one. It’s the sound of your own voice bouncing off hardwood floors and empty walls. If your office is a small, boxy room with hard surfaces, your voice pings around like a rubber ball. This creates “resonance” (a ringing sound at specific frequencies) and “reverb” (the hollow, echoey sound).

Why this matters: In real life, the sound of a room is always present. In fact, if you were in a truly “dead” room (an anechoic chamber), it would feel surreal and unsettling. When you are speaking face-to-face, your brain uses spatial cues to automatically filter out those reflections. It “knows” what the room sounds like and subtracts it so you can focus on the voice.

The Remote Problem: When you are on a Zoom call, your listener’s brain loses all those natural cues. It doesn’t know you are in a tiled kitchen; it just hears a wash of muddy, distant noise. Without those cues, the brain can’t do its filtering work, and you end up sounding unclear, distant, and fatiguing to listen to.

The Solution: When you use software to remove that room reverb, you are essentially doing the work that the brain would normally do.

You might think that removing all room sound would make you sound unnatural, but the opposite is true. We have been trained by decades of movies, radio, and high-end podcasts to associate “dry,” direct audio with authority and professionalism. By stripping away the room tone, you don’t just sound clearer; you sound like a broadcaster.

If you can remove the sound of the room, do it.

The three weapons of silence

To fight these noises, we have three main categories of tools. You might need one, or you might need all three.

1. The Gate (The Sledgehammer)

We covered this in depth in Part 2 , but it’s worth a quick recap. A gate is just an automated volume knob that turns off the sound completely when you stop talking.

The downside? It doesn’t actually remove noise; it just hides it when you aren’t speaking. When you talk, the gate opens, and the air conditioner noise comes rushing back in underneath your voice. It’s great for “black silences” between sentences, but it won’t make your voice sound cleaner.

Pro Tip: You can actually use a gate to fight dynamic noise, like a noisy keyboard or high-strung dog (yes, I’m talking about you, Kona!), by using the side-chain filter. As we discussed in Part 2, most advanced gates let you filter what the gate “hears.” If you have a high-pitched noise, you can filter the high frequencies out of the side-chain. This tells the gate, “Ignore high-pitched sounds.” I use this technique frequently if I’m working near my daughter, since her voice is a lot higher-pitched than mine. She might still be audible while you talk, but they won’t trigger the gate to open when you are silent.

2. Spectral subtraction (The Scalpel)

These are the traditional “De-Noise” plugins (like Bertom Denoiser or iZotope RX Voice De-noise). They work by looking at the frequency spectrum of your audio (low bass to high treble) and lowering the volume of only the frequencies where the noise lives.

The workflow usually involves “teaching” the plugin. You highlight a section of “silence” where you aren’t talking, and the computer learns the profile of the hum or hiss.

Good for: Constant hums, fans, and A/C units. Bad for: Room Reverb. Why? Because the “noise” (the echo) only happens while you are speaking. It doesn’t exist in the silence, so the plugin can’t learn to remove it. You can sometimes use EQ to surgically cut resonant frequencies, but that is a fragile fix—if you move your mic three inches, the resonance changes, and you have to start over.

Tools to try:

• Bertom Denoiser: This is hands-down the best free option on the market. It’s a “zero latency” plugin, meaning you can use it live on calls without delay. It uses a multiband approach (think of it like 5 gates working on different parts of the sound) to push down the hiss without killing your voice.
• iZotope RX Voice De-noise: The industry standard. If you watch a movie or listen to a high-end podcast, they probably used this. It has an “Adaptive Mode” that listens to the noise floor change in real-time and adjusts itself automatically. It is incredibly transparent but usually costs money.
• Klevgrand Brusfri: This is a fascinating middle ground. It essentially uses multiband gating to do the job of a spectral denoiser. You hold a “Learn” button for two seconds of silence, and it builds a profile to gate out the noise frequencies. It is incredibly effective for static noise.

3. AI isolation (The Magic Wand)

This is the modern era. Tools like Supertone Clear , Waves Clarity Vx, or Krisp don’t really “subtract” noise in the traditional sense. They use machine learning to identify what a human voice is and rebuild the signal containing only that voice.

They don’t care if the noise is static, dynamic, or reflective. They just know “This frequency is a human vowel” and “That frequency is a wall reflection,” and they delete the reflection.

Good for: Everything, especially room reverb. This is the only tool that can effectively “dry out” a wet, echoey room. The downside: They are CPU heavy. Running these in real-time on an older laptop can cause glitches or latency. Warning: These are voice only. Because the AI is trained specifically on human speech, it will interpret a guitar, a synthesizer, or a melody as “noise” and try to destroy it. If you do any music production, do not use these tools on your instruments.

The controls: How to drive the car

Most noise reduction plugins (especially the Spectral ones like Bertom or RX) share the same basic controls. Here is what they actually do.

Threshold

This is the most important knob. The Threshold tells the plugin, “Anything louder than this is a human; anything quieter than this is noise.”

• If you set it too high: The plugin thinks everything is noise. It will start eating your voice, making you sound muffled or cutting off the ends of your words.
• If you set it too low: The plugin thinks the noise is part of your voice, and it won’t remove anything.
• How to set it: Raise the threshold until the hiss disappears, then stop. If your voice starts sounding underwater, back it off a little.

Reduction (or “Amount”)

This controls how much volume to take away from the noise once it’s detected.

• 0dB: The plugin does nothing.
• -6dB: The noise is turned down by half. This usually sounds very natural.
• -100dB (or Max): The plugin tries to delete the noise entirely. This is where you get those robotic artifacts.

Attack and Release

Just like the Gate we discussed in Part 2, these control the speed of the robot.

• Attack: How fast the reduction kicks in when noise is detected.
• Release: How long it holds onto the reduction after the noise stops.
• Pro Tip: For voice, you generally want a faster release. If the release is too slow, the “suppression” will hang around too long and might accidentally clamp down on the start of your next word.

The “Roam Office” strategy

So, which one do you use? If you are just sitting in a quiet home office, a simple Gate might be enough. But for travel, I use a specific hierarchy of noise reduction.

Physics first

Before you load a single plugin, remember the Golden Rule of Audio: It is easier to prevent noise than to fix it.

Software can do magic, but magic comes at a cost (usually in audio quality). If you are in a noisy hotel room, do these four things before you open your laptop:

1. Kill the AC: I know, it gets hot. But turn off the HVAC for the 30 minutes you are recording or presenting. That is 10x better than any plugin.
2. Pillow Fort (and Blankets): If the room is echoey (which makes noise reduction work harder), put pillows behind your laptop and sit facing a bed or curtains. Better yet, hang a heavy blanket behind you or over the door. Soft surfaces absorb noise; hard surfaces reflect it.
3. Choose Your Weapon: If you have the option, use a dynamic microphone rather than a Condenser mic. Dynamic mics are naturally less sensitive to high-frequency transients and distant room noise. Also, check your polar pattern. Ensure you are using a cardioid or super-cardioid pattern. These act like a flashlight beam, picking up sound only from directly in front of them while rejecting the keyboard clacking and room reflections coming from the sides and rear.
4. Proximity Effect: Get the microphone closer to your mouth. We talked about this in Part 1. The closer you are, the less “gain” you need, and the less background noise the mic will pick up relative to your voice.

The order of operations

It’s not enough to just have the plugins; you have to put them in the right order. In audio, the “Signal Chain” flows from top to bottom.

The Golden Rule: Noise reduction should happen as close to the source as possible.

1. The Source (Microphone): If your microphone (or audio interface) has a built-in “Low Cut” or noise reduction switch, use that first. Hardware reduction is almost always cleaner than software.
2. Noise Reduction Plugin: This must be the very first plugin in your software chain. You want the AI or Spectral tool to hear the raw, dirty signal so it can clean it. If you put it later, you might accidentally compress the noise, making it harder to remove. Or worse, putting it after a gate could prevent it from even knowing what noise to remove.
3. Corrective EQ: If you need to cut out a specific ringing frequency (resonance), do it here.
4. The Gate: This comes last. You want the gate to react to the clean signal. If you put the gate first, it might struggle to distinguish between your voice and the A/C unit.

Trust your eyes (Since you can’t trust your ears)

In a studio, engineers listen to their changes in real-time. But in a Zoom call, you can’t really do that. The latency (delay) introduced by the plugins makes it impossible to speak while listening to your own processed voice—you’ll end up slurring your words.

This means you are effectively flying blind. If your noise reduction glitches out and starts eating your words, you won’t know until someone says, “Hey, you sound like a robot.”

The Fix: Visualization. When I set up my DAW for meetings, I always keep a spectrum analyzer open. I configure it to show two lines:

1. The Raw Input: What the mic hears (usually a messy line with lots of low-end rumble).
2. The Processed Output: What the team hears (a clean, tight line).

I arrange my windows so this graph is always visible in the corner of my screen. By keeping an eye on it, I can visually confirm that the “noise” (the flat line at the bottom) is disappearing when I stop talking, and that my voice frequencies aren’t getting chopped off.

Spectrum analyzer showing just the the noise-removal step

Spectrum analyzer showing the overall processing chain

Spectrum analyzer showing raw (blue) vs processed signal (orange)

I actually typically set up a couple of these. One for my overall processing chain, and another for just noise removal stage. Above, you can see the affect of Supertone Clear on my signal. Notably, I’m not actually speaking there; it’s actually picking up the sound of a TV nearby, since my mic is laying on its side and is pointed across the room. The polar pattern of my microphone, combined with a gate, should clean that up nicely. And more than likely, for a meeting, I’d just go to another room to get further away.

How to set up dual-spectrum before/after monitoring in Bitwig Studio

1. Add a “spectrum” device before your noise removal plugin.
2. Click on it (as in, click the word “SPECTRUM” in the device panel), to see its options in the info panel.
3. Click on the grayed out B button in the info panel, to enable the second input.
4. Click on the “Device Input” button next to the B button, then browse to your noise removal plugin.

The plain spectrum analyzer

The analyzer’s info panel

Info panel with the second input enabled and configured

The end result

Setting up a before/after spectrum analyzer

The artifact danger zone

Here is the trap everyone falls into: You get a new noise reduction plugin, crank the knob to 100%, and marvel at the silence.

But then you listen to your voice.

When you use heavy noise reduction, you get artifacts.

• The “Underwater” Sound: Your high frequencies get garbled. You sound like you are talking through a bubbly aquarium.
• The “Gate Chops”: If settings are too aggressive, the ends of your words (like the “t” in “cat” or the “s” in “yes”) might get cut off.

My Advice: It is better to have a little bit of background hiss and a natural-sounding voice than to have “perfect silence” and sound like a robot. Aim for 80% reduction, not 100%. People’s brains will tune out a little hiss, but they will be distracted by a robotic voice.

My Personal Recommendations

I have tested dozens of these tools in the field, from airport lounges to quiet home offices. If you want to cut to the chase, these are the two I actually use:

1. The “Clean Up the Hiss” Choice: Klevgrand Brusfri I mentioned this in the spectral section, but it deserves a spotlight. I previously reviewed this plugin here , and it remains my go-to for standard noise. Because it uses advanced multiband gating rather than aggressive spectral subtraction, I find it leaves the “tone” of my voice alone more than other plugins. It’s lightweight, fast, and handles A/C hum perfectly.

Bonus: Because it isn’t an AI model trained on human speech, it works beautifully on other audio sources. If you do any music production and want to clean up a noisy guitar amp or synthesizer recording, this is the tool to buy.

2. The “Nuclear Option”: Supertone Clear When the environment is truly chaotic—dogs, traffic, or a massive room echo—this is the winner. I reviewed Supertone Clear recently here , and its AI engine is magic. It separates the voice from the ambience (and even room reverb) better than anything else I’ve used. It’s heavier on the CPU, but when you absolutely need to kill the background, this is the tool. In an echoey noise or unpredictable/dynamic environment (so… most of the time for me, since I travel so much), I go with Clear.

What’s next

At this point, you have a good mic, you’ve gated the silence, and you’ve scrubbed the noise. You are clean. But you might still sound a little… thin. Or maybe quiet. Or maybe inconsistent.

In the next post, we are going to talk about compression and limiting—the art of making your voice sound loud, authoritative, and consistent, without blowing out everyone’s speakers when you laugh.

Stay tuned for Part 4!