(EDIT: Found out that I definitely didn't invent this. Heart breaks (celery snap + cold spaghetti sauce squish + the sound of a tiny, tiny violin))

Lately it seems that a lot of people have been asking about how to remove reverb from location dialogue, and I've been dealing with it myself on a current project. Because I have no life, I've been sitting (on a Saturday night) playing with the dialogue, trying to see if I can get rid of this ass ton of horrible noise. I will be heart-broken if someone has already come up with this, but I think I've just invented a Multi-Band Gate:

Routing Setup for Multi-Band Expander http://1.bp.blogspot.com/-73NdQu2q9lA/TZexAQ8gV3I/AAAAAAAAAB0/Hy6p4XweyHU/s1600/016.JPG

In the example above you will see that I have 4 compressors and only one gate. Explanation follows:

Of late I've been finding many references to multi-band compression as a great noise removal tool if you don't have fancy NR plugs like iZotope RX. So, I thought I'd give it a try. Sadly, I don't own a multi-band compressor, so I thought I'd make one. The principle is really pretty simple, here's how you do it:

Take your original signal and route it into 4 (or more if you like) separate Aux Channels. In the inserts drop in an EQ7 (or anything else with a top and bottom filter in it), and slice it up into it's component bands. In this case I've done Low: 0-120Hz, Low-Mid 120-1000Hz, Mid: 1,000-7,500Hz, High: 7.5kHz-20kHz. For the sticklers out there, I know that this is not precisely how the bands are broken up, but this is how my brain thinks they should be (don't ask me, he's a really weird dude). So, I picked numbers that split the spectrum up nicely. You can set it pretty much however you like, same goes for your slopes.

Once you're done with that, all 4 Auxiliary tracks are routed back into a single track, thereby giving a single overall volume control. The ultimate aim is to have the 4 filtered tracks come back sounding exactly the same as the single original.

Once you have your bands and crossovers playing nice together you can drop in your compressors. So, in stead of compressing the entire signal, you can pick a part of the signal and compress that. In this case, I tried it on all 4 bands and it just wasn't working, if anything it was making the reverb worse. Naturally, I proceeded to get really frustrated and irritated with myself for wasting so much time on a stupid idea, until...

My brain went sideways and remembered that a lot of people like to use gates/expanders when faced with large amounts of noise or room. So I dropped in a gate on one of the bands. For me, the most irritating and noticeable part of a big reverb tail is the mid-ish band 1-5kHz (I suspect this will be true for most people) so that's where I put my gate. My band as-set was a little wider than that, but I was more interested to see if it would actually work than being absolute in my bandwidth accuracy.

Now, I generally find gates used on dialogue to be really choppy and obvious, unless it's done spectacularly (a feat which I will readily admit I've never accomplished). It's the high-end hiss jumping in and out that gives it away. However, in this case, because it's only gating 1-7.5k, all of the high-end hiss is left exactly where it was in the first place. This means that you can set your gate to really cut hard, and the rest of the frequency content of your signal will cover your tracks for you. This is doubly true if your content is as F*^*ING noisy as mine is.

And I have to say, it really works well. There are three great things about it:

One: The reverb is most definitely still there, but it doesn't eat into the dialogue in the way that it used to.

Two: You're not losing the Mid content of the dialogue entirely, just the tail, so the words themselves actually end up sounding pretty close to the original.

Three: Like the reverb, the part of the spectrum that gives ambient noise its beef is the 1-5k range, so this method controls a lot of that too.

Bonus Prize!: You can use the remaining 5 EQ bands in the EQ7s to shape each band of your multi-band gated signal. This quadruples the number of bands you have to use on the entire thing. You've just built yourself a 20-band paragraphic EQ.

As you can see I'd already dropped a compressor onto each band, so I used it on the high and low ends to control some of the hiss and rumble, and everything sounds way less terrible than it used to.

Oh, and another really cool thing... this method also gives you large-scale faders for controlling the volume of each band, in stead of the fiddly little knobs in the EQs themselves.

I am totally high-fiving myself right now.

Your Wish = My Command

Before: [soundcloud]g-a-harry/before-side-chain-gate[/soundcloud]

After: [soundcloud]g-a-harry/after-side-chain-gate[/soundcloud]

Still needs a touch-up. But now I actually have something to work with.

  • Nice work! I do that for far out voice processing sometimes. Now post a before and after file. :)
    – Rene
    Apr 3, 2011 at 4:27
  • @Rene, apparently SoundCloud hates me right now. I posted a file in another thread and it gave me the same "not available" guff.
    – g.a.harry
    Apr 3, 2011 at 5:05
  • Great post! Keep em coming!
    – Chris
    Apr 3, 2011 at 20:07
  • This workflow is very simply with McDSPs ML4000 muliband limiter/compressor/expander/gate. Worth checking it out, sounds great. Apr 3, 2011 at 21:15
  • 1
    SPL make a simple RTAS plugin called De-Verb, does quite a respectable job for very little money.
    – ianjpalmer
    Apr 4, 2011 at 8:30

1 Answer 1


Ever try this with the Waves C4? Been using the method for a few years now, sounds similar to what you mentioned. I did a blog writeup on it a few months back for dialogue noise, it may be of interest as another means of reaching a similar solution:


  • 1
    Great post, man.
    – lucafusi
    Jan 25, 2012 at 6:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.