How to Remove Background Noise from Voice Recordings Pt.3

We've already learned how to recognize background noise and what makes the spectral de-noiser in iZotope RX so powerful.

Now we'll learn how to train the spectral de-noiser to allow it to remove background noise in your dialogue recordings most effectively.

This method will set you up for crisp and clear sounding voice recordings for your listeners, free of distracting background noise!

Teaching the Spectral De-noise Module

The spectral de-noiser works by using a custom created noise profile specifically for your voice recording.

However, you have to teach the spectral de-noiser so it knows how to create your handmade (and therefore extremely effective) noise profile.

Because of this, it would be a good habit to start recording 5 seconds of "dead air" on all your dialogue recordings for the de-noising process.

"Dead air" means recording with no speaking or it means recording just the sound of the room.

What's the reason for this grim sounding "dead air"?

The "dead air" in your recordings will provide the information to teach the de-noiser.

You could record this 5 seconds of "dead air" at the beginning or end of your recordings.

This practice of recording the sound of the room is important for creating the most accurate noise profile.

And when you're able to do that, the de-noise module will do its best work.

It will reduce the background noise during and between phrases while leaving your dialogue mostly untouched!

Wait, I Don't Have 5 Seconds of Dead Air...

While 5 seconds of "dead air" is considered best practice, you may not have that available on your dialogue recording...

Maybe you've never heard of recording the sound of your room for voice recordings or maybe you have a great take and just forgot to leave some "dead air".

No worries, you don't need to re-record that fantastic take!

Let me show you a tip to train the spectral de-noiser with a lot less than 5 seconds of noise:

• look at the image below from my actual voice-over project...
  
  
• the selected area is only persistent background noise (nothing else such as breaths etc.)...
  
  
• the blue waveforms are my recorded dialogue...
  
  
• notice the seconds listed at the bottom of the image...
  
  
• imagine moving the selected area a little to the left to 0:57.0...
  
  
• you can see that it would about fill the area from 0:57.0 to 0:57.5 (1/2 of a second)...
  
  
• this means I'm using just 1/2 of a second of background noise between phrases to train the spectral de-noiser!

There are several reasons for this working with such a short selection of noise.

First, I made sure to use only persistent or constant background noise in my short 1/2 second selection (this is important and I'll tell you why in the next section below).

Next, I tested out the sound of the spectral de-noiser using this selection as a noise profile.

This means I listened through the voice recording while turning the de-noiser "on" and "off" to see how it affected my spoken words.

I flipped back and forth between having the spectral de-noiser engaged or bypassed listening to make sure the dialogue still sounded natural and clear.

If I had noticed the de-noiser negatively affecting the sound of my voice, that would have been a red flag prompting me to go back and change something...

In fact, it sounded like a huge improvement as it removed general noise and hiss throughout the voice recording leaving clear sounding dialogue!

So there you have it - fantastic noise reduction using just 0.5 seconds of pure background noise.

You can do the same thing on your dialogue recordings today!

Which Noise Do You Use?

As you're going through your voice recording looking for your 1/2 second of background noise, you'll probably run into a variety of noises.

Hiss, low rumbles, clicks, pops, breathing sounds, mouth smacks, hums etc.

But which one should you select for RX's spectral de-noiser and which should be left alone for a different removal process?

What has worked for me is to use only consistent or perpetual background noise.

You'll probably need to turn up the volume and listen carefully to pick out the noise in the background. Plus, it will usually occur at a much lower volume level than your recorded words.

Then, make sure the noise you're focusing on with your listening is constant or ongoing through a lot of or all of your dialogue recording.

This is what you want to select to train the spectral de-noise module.

That means no breaths or mouth clicks (which are temporary noise), but yes to constant hiss, rumble and hum.

For example, in my voice-over project:

• when I tried selecting background noise that included both constant hiss AND me taking a breath, it didn't work...
  
  
• the frequencies of my breath were added to the noise profile of the hiss and lower frequency noise...
  
  
• instead of removing breath sounds, the de-noiser wanted to remove frequencies of my voice...
  
  
• these frequencies of my voice were those that corresponded to the breath frequencies I selected above...
  
  
• so it was removing parts of my voice along with the background hiss and lower frequency noise - not good...
  
  
• INSTEAD, you'll want to select only the background noise between phrases as I've done in the image below.

Below you'll see the same image from earlier in this post:

• the small red box and arrow designates a mouth noise at the end of a phrase that you don't want to add to your background noise selection...
  
  
• the large red box and arrow designates me taking a breath before a spoken phrase (you don't want this in your selection either).

Summary

Remember, the best way to teach the spectral de-noise module is to record around 5 seconds of the sound of your room.

This "dead air" is important for creating the most accurate noise profile.

Next, don't panic if you don't have 5 seconds of "dead air" in your voice recording.

You can get great results using as little as 0.5 seconds of consistent background noise found in pauses or between phrases of your recording. 

Finally, make sure you use only consistent or perpetual background noise to train the de-noiser.

This will result in crisp and clear sounding dialogue with greatly reduced noise that your listeners will enjoy!

Have YOU used these steps on a recent dialogue recording? How did they help?

Would love to hear your thoughts below!

Cheers, Bryan
Your Dialogue Editing Mentor