Audacity Tutorial – Noise Reduction (formerly Noise Removal)

Posted August 22, 2015 by Shaun

Audacity is a powerful, free audio editor that’s been indispensable to podcasters since the inception of the medium. For someone new to audio production, however, it can be confusing and intimidating. This series of tutorials is aimed at the podcaster or amateur broadcaster who doesn’t want their great content obscured by bad audio quality.

Please note that these tutorials don’t avoid jargon. It’s everywhere in Audacity itself, so avoiding it seems counterproductive. Instead, I try to define as many terms as possible from the perspective of someone with little- to no background in audio production.

Tutorials in this series

» Compression using the Chris’s Dynamic Compressor plugin
» Noise Reduction (formerly Noise Removal)
» My Plug-Ins Aren’t Showing Up!
» Common Usage Scenarios for Noise Gate (currently in progress)
» Useful Equalization Presets (currently in progress)
» Increasing Levels, Avoiding Clipping with Peak Limiter (currently in progress)
» Suggested workflow: Poor-quality original recordings (currently in progress)
» Suggested workflow: Mid- to high-quality original recordings (currently in progress)


Introduction

The first step to reducing noise in Audacity is telling the application exactly what you consider “noise,” so that it doesn’t reduce or remove any actual content. This is discussed in the first section, Creating a Noise Profile.

The second section deals with noise reduction itself, with detailed descriptions of and situation-specific recommended settings for the Noise Reduction effect.

Lastly, because Noise Reduction will never produce a perfect result, manual removal of stray noises is often necessary. The final section discusses manual noise removal techniques.


Creating a Noise Profile

nerd3-1

The first step in noise reduction is to give Audacity a sample of the noise that should be reduced. I’ve found that around 20 seconds of noise sample is best. For this reason I highly suggest that, when you record, you record “silence” (which is actually ambient room noise) for 20 seconds before you start speaking. This will give you a great, clean noise sample.

If you’ve got a clean, contiguous noise sample, great! You can skip to the next section.

Don’t panic, however, if you don’t have 20 consecutive seconds of noise sampled. Chances are, you’ve got at least 20 seconds of accidental ambient noise scattered throughout your recording. Let’s look at how to combine some of those into something usable.

NOTE: Do not copy the same noise sample multiple times to fill out your 20 seconds! 20 copies of the same 1-second sample will produce the same result as just using one 1-second sample.

nerd4-1

Here, I’ve selected just under 2 seconds of noise between speaking segments. Notice the visual contrast between background noise (“jagged” looking) and speaking (“smoother,” more “solid” appearance). I’m going to comb through the clip, find similar-looking stretches of background noise, and copy as many of these as possible onto a new track.

Tracks are added under the Tracks menu, under “Add New.” Be sure to add a track with the same number of channels as your source material. For example, if your source is stereo, use “Add New > Stereo Track.” If your source is mono, use “Add New > Audio Track.”

nerd9-1

In the image above, the track at the top is the original clip, from which I copied noise. At the bottom is the newly-created track, into which I pasted scattered bits of noise from throughout the original. Once I gathered all of the noise chunks I needed, I used the Time Shift Tool to slide them together into one super-chunk of noise.

Because all of these instances are separate copy/paste jobs, Audacity still treats them as separate, as indicated by the black lines between each clip. The join function- found in the Edit menu at Clip Boundaries > Join- removes these markers and consolidates all of these fragments into one big piece.

nerd10-1

At the bottom you see the block of noise I’m going to work with, after using the join function. My selection of noise was a bit inconsistent, as you can see by the appearance of some straggler peaks.

The more exact the noise sample, the better Audacity will do at reducing only the noise. Therefore, it’s a good idea to remove things like inhales, exhales, and clicks.

Remove the peaks and other unwanted anomalous noises by dragging your cursor to select each piece of anomalous noise and then hitting the delete key.

nerd11-1

Here’s what my noise sample looks like with anomalous noises deleted. Compare this image to the above, and notice that it is far more uniform. It’s easiest to see the difference by looking at each image’s peaks.


Noise Reduction

Armed with a pretty good noise sample, it’s now time to provide that sample to the noise reduction effect. Select the noise block created in the previous step, then go to Effects > Noise Reduction. A window will pop up; under Step 1, select “Get Noise Profile.” Note that no settings need to be adjusted at this point, even if your settings differ from those shown in my screenshot.

nerd12-6

A bit confusingly, you won’t be given any feedback to confirm that this operation has been completed successfully; instead, the window simply disappears. But don’t worry: it’s been grabbed!

Now that Audacity knows what I consider to be “noise,” I can apply noise reduction to the clip. Select the clip that you wish to remove noise from, and then reopen the Noise Reduction effect (Effects > Noise Reduction).

nerd13-4

Now, because we’re at Step 2, we’ve got some settings to deal with. Here’s what they mean, very generally:

Noise Reduction (dB) TL;DR VERSION: Start at 30 dB and adjust as necessary. Keep as low as possible.
This setting controls how many dB’s Audacity will reduce your noise by. Typically 30 dB is a good place to start. If in a controlled recording environment, a lower number like 10 or 15 dB might work. In a noisier situation, try a higher number like 35 or 40 dB.

It’s good practice to keep this number as low as possible, because Audacity attempts to remove noise from the entire track, including spoken sections. The higher the number, the more likely it will be that Audacity will reduce non-noise sounds- like speaking- from your content.

Sensitivity (dB) TL;DR VERSION: Start at 4.00 dB and adjust as necessary. Keep as low as possible.
This setting tells Audacity how aggressively to remove the noise, even at the risk of removing audio that isn’t noise. I like to think of this number as being equal to the size of a bull in a china shop. In this example, the china shop is the “good” parts of your recording, so even though you’ll probably break some amount of china by removing noise, you want to keep the damage (and therefore the bull) as small as possible.

As with the previous setting, keep this number as low as possible. Sometimes I’m able to get away with 0.00 dB, but I’ve also had to go as high as 10 dB. If you find yourself consistently using a number higher than 2 dB, it’s probably because you aren’t close enough to your microphone, resulting in a bad signal to noise ratio. Get closer to your mic!

Frequency smoothing (Hz) TL;DR VERSION: Just always use 6.
This setting gets a bit technical, so I’ll skip an involved explanation of what it does. If you’re curious, check out the official Audacity manual’s page on Noise Reduction.

As far as what number is best, that depends. Frequency smoothing can mitigate “sparkly” artifacts resulting from noise reduction. Most people have heard these kinds of artifacts on low bitrate mp3s. If you hear these artifacts in your file, bump the number up until they’re either gone or gone enough so that you don’t notice them.


I almost never get the settings right the first time around, so don’t be surprised if you find yourself in a classic undo-adjust-reapply-undo cycle. Here’s what it looked like on my first try:

nerd14-2-1

Eyeballing it, this actually looks pretty good. As a reminder, here’s what the clip originally looked like:

nerd14-3-1

You can see, especially at the beginning (far left) of the clip, that Audacity has removed a significant amount of the persistent background noise.


Manual Noise Removal

Even though the ambient noise is pretty well removed from this clip, some unwanted sounds remain. Instances of loud breathing, plosives, lip smacking, and keyboard clicks are all common, because these sounds have completely different profiles than the ambient noise.

There are two quick ways to handle these unwated sounds: deamplification and silence. As pure silence can be jarring, deamplification is usually preferable.

nerd20-1

Here I’ve identified a mysterious noise (I think it’s a lip smack) that’s more distracting than useful.

Getting rid of this noise is easy: select the unwanted noise and either silence (Generate menu > Silence…) or deamplify (Effect menu > Amplify…, enter a negative number; -60 dB or so should do it). Below, you’ll see that the noise is now gone.

nerd21-1


That’s it for basic noise removal. If there’s anything else in the post production or editing processes that you’d like me to cover, please feel free to leave a comment.

If, like a lot of podcasters, you’re too busy generating content than to mess with Audacity settings, get in touch. For a nominal fee, I’m happy to complete a round of post production and provide a detailed write-up of what I did, so you can do it yourself from then on.


← Previous: Audacity Tutorial – Compression Using the Chris’s Dynamic Compressor Plugin

Next: Audacity Tutorial – My Plug-Ins Aren’t Showing Up! →