Anybody got any tips on performing noise reduction without using any of the fancy NR tools (iZotope RX, X-Noise, et. al.?). I'm one of those youngsters who has never really had to work without them, so I'm finding myself a bit flummoxed.

I'm working on a project whose primary mandate is authenticity of aspect, nothing canned, and everything as raw and real as possible. Apparently raw + real = poorly recorded, but I won't start into that. Now, I know for a fact that if I leave everything as RawReal as possible (read, do nothing to the location dialogue at all) I'm going to get a moon high stack of notes from everyone involved. So I thought I'd kick it old school and work some sushi chef magic with tight Q exacto knives.

What that means, of course, is that I've been sitting here all day playing with multiple instances of the 7-band EQ3 and the Expander/Gate Dyn 3, chopping up great swaths of room tone and mucking about with side-chains. But I just can't seem to find a setup that works for me.

The tack I've been trying to take is not to remove all of the noise, just the annoying bits, 3.2k, 15k, 6.5k, 2.5k, etc. But it all ends up sounding like they're talking into a mattress...

Any ideas on places to start? Bands to look at? thresholds and expander ratios that might work?

I'll use RX if I have to, but it'd feel like cheating.

  • I'd rather use RX with gentle settings rather than the phase flipping trick. I wouldn't consider it cheating, RX stands for one of the tools that have made audio work accessible to people on a budget like us and you're literally shooting yourself in the leg by refusing to use it... It sounds like you've used a pot holder to move a pot away from the burner but now you decide to manipulate it with your bare hands... Any frequency band you'll want to keep (for dialogue, pretty much everything from 80 to 16k) you'll have to stick with the noise in it, unless you use an intelligent process... my 2c Commented Apr 3, 2011 at 1:19
  • 1
    @g.a.harry, well, at first you duplicate and flip so it plainly cancels itself out, let's call the duplicate a "mask". Then you EQ out, in your mask track, the frequencies that you actually want to hear, this will "unmask" the same frequencies in your original recording. To be honest, I find this method quite harmful for the recording, maybe I could just never get it right... Ideally your mask would only contain the background noise to cancel it perfectly, but that is just not possible. Commented Apr 3, 2011 at 14:35
  • 1
    @Justin Out of curiosity, how would the phase flip/eq differ from simply cutting the signal by XdB before the eq, and boosting the same frequencies you'd be cutting for the phase inverted track? In theory, it seems to be exactly the same, but then theory doesn't always translate to practice... Commented Apr 4, 2011 at 4:23
  • 1
    @Roger, I had the exact same question and my answer is that there is no difference. I believe @Iain McGregor mentioned this trick was better executed with your main and mask tracks being simultaneously recorded by two matched mics physically very close (to prevent phasing, as we know). Then the masking technique makes more sense. In the duplicate method, I think it merely acts as an interface shift, with which you start with zero noise/sound and then build it up. This doesn't sound good to me since you're not listening to the original recording in its whole... Commented Apr 4, 2011 at 22:05
  • 1
    @g.a.harry I believe it yields a more analogue feel to the technique. Duplicating and flipping the phase sounds like a nasty plain cancellation, whereas using two distinct recordings will never get you the absolute zero and it would probably be for the best! Commented Apr 8, 2011 at 20:37

5 Answers 5


Just a quick note: i've found that, for dialogue, if you use a narrowish Q (around 3 or 4?) in the vicinity of 800Hz-1kHz, you can lower a bit of the noise without interfering too much with the signal.

My 2 cents on the phase inversion stuff: not worth the time. In order to cancel out your noise, you'd pretty much need to record it from the exact same angle, with the exact same mic, at exactly the same time as the noise in your recording, etc. If there's even a slight difference in the phase inverted noise recording, you'll just be adding to the noise in your dialogue. Maybe, if you work really hard on it, you could weaken the noise a little, but the chances of getting it right are very slim.

I'm happy to be educated on this, but this is the conclusion i came to when i had a similar problem way back in the pre-RX days.

  • @Roger, even recording from the exact same place, angle, etc. won't work. You have to use the exact same file so you have the exact same waveform, otherwise, yes... all you're doing is adding noise. The problem then is minimized to only when they characters are talking ('cause obviously you don't want to cancel that out.) Commented Apr 4, 2011 at 21:48
  • @Syndicate that's what i meant by "at exactly the same time", and it obviously wouldn't work, because the character is speaking at that time. Commented Apr 4, 2011 at 23:11

Multiband expansion. WaveArts Multidynamics, McDSPs ML4000, or Waves C4/C6 do this.

Use EQ to remove offending room modes, then add multiband dynamics gently..


Using the tools you have is far from cheating, however I do appreciate your interest in learning alternate methods.

Personally, I've never had to really kick it old school with noise removal with the exception of experimenting and doing what you're doing and experimenting with EQ and Gain. One trick I've been able to use from time to time is canceling out the noise via phase inversion trickery and EQ filters. This is a little tricky to get working properly but it is possible depending on what the source of and how cyclic your ambient noise is.

  • @Syndicate, I've tried the trixy phase stuff, but could never get it to work. But most of the time I'm dealing with really broadband stuff. It's essentially pink noise with some dialogue thrown in for good measure.
    – g.a.harry
    Commented Apr 3, 2011 at 0:04

Well I just finished mixing a feature using Cedar DNS, Izotope RX, Waves C4 and tons of eq automation and it still sounds raw due to the nature of the production sound. Far from cheating IMHO. If the mixers of Citizen Kane had this gear they would of used it too.

Plus one for multiband expansion in your case, like Waves C4 or WaveArts Multidynamics. Besides those just emulate what the old school gear (I.E. Dolby CAt 43) did anyway.

Good luck.


@G.A.harry New school is the New old school!

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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