Noise Reduction, a Judgement Call

I’m about to embark on a noise reduction job consisting primarily of Mic Preamp hiss. Broad band. The worst kind.

I am also about to metaphorically write a check to Brian Davies for his N/R package. I have been able to achieve very good noise reduction (in other jobs – and using all the tricks), but have never been able to get rid of noise that occurs during spoken words. So the show (if it’s bad enough) always ends up sounding like a space alien on the snake or amphibian model. Hissy words.

That’s hopeless, right? No software package can fix that. I think Brian’s tutorial’s suggest something of the kind.


Using the standard (not the latest) Audacity noise reduction, I’ve found that boosting* the high frequencies of the “noise only” sample before obtaining a “noise profile” from it improves the result: proportionally more high frequencies removed => less snaky sibilance.

Mr Davies doesn’t seem to be applying noise reduction equally to all frequencies either …

[* this sort of boost …
boost given to  high frequencies before obtaining 'noise profile'.jpg
a more complex (psychoacoustic) equalization curve, rather than this straight line, may give better results ]

Our old friends Fletch and what’s-his-name. That makes sense. Certainly noise at 3KHz will drill holes in your ears compared to noise in other places. I hadn’t thought about configuring the profile…

And what about the tools in Audacity Alpha…since you mentioned it.


I haven’t tried it yet, last I heard (September) it still had some bug in the controls …

OK. All this haystack, Fletcher-Munson, response tailoring stuff is lovely, but at the end of the day, isn’t it true that noise correction is not applied to voices? In fact, success in noise reduction depends largely on recognizing when someone is speaking and not applying any correction at all – the better to prevent space alien effects.

This was the problem with the noise reduction in Audacity 1.2. It couldn’t tell when someone was speaking and the desirable correction adjustment was always the smallest fraction up from completely off.

Frequency Smoothing and Attack/Decay are chosen to gracefully suppress voice distortion, not amount of noise reduction in the rest of the show. That’s set by dB Reduction like it always was.


IMO It’s necessary to apply noise reduction to all of it : words and gaps, then apply a gate to cut back the volume of the hiss in the gaps between words. Before writing your cheque make sure the software can actually produce a better result than Audacity, I’ve not tried Mr Davies program but I have tried trial versions of GoldWave and others, they all create artifacts when Noise Reduction is applied …

[ Steve’s noise gate here … ]

I love the clip. A terrific announcing voice that in real life sounds like he’s from North Carolina.

“Dammit. Ah knew ah was going to say that.” It’s not that pronounced, but I can still get him to one or two states no trouble.

It gets much more serious during musical performances. There you can’t have very many gates running during a song. I wonder how well I can do with a heavily processed profile and almost no frequency smoothing and very slight suppression values.

What can you do with performances in stereo where there is only one singer, but the noise is in stereo?


I stumbled upon this in my search for answers to the noise issue I’m currently dealing with. Would be much easier to build a time machine and record it correctly in the first place (probably quicker too :smiley: ).

Anyway gating might help me out a little here too - so thanks for reminding me :bulb: I’m sure I thought about it last night at some stage. Was a verrrrry late one.

Anywhoo to answer your question Koz - to remove stereo noise from a track recorded with presumably 1 microphone - bounce it to mono, use your gates/noise reduction tools etc then bounce it back to stereo. Or am I missing the bigger picture?

Well anyway now I’ve removed my ground loop hum and variants of it - I’m pretty much left with a lovely white noise hiss :blush: . I feel ripped off~ If anyone has any tips on removing that without my beatboxer sounding like an alien or he’s been recorded inside a duvet full of sand - please let me know!!! He wants me to put it on the web and sell it as a live recording… :open_mouth:

Maybe you could put his beatbox vocal through a strong effect to disguise the hiss.
This “bitspeak” effect I’m told is the latest thing … (but I’m so old it sounds retro to me),
There is also the Cher/ T-pain extreme autotune effect … (but that effect is a bit old hat).

Thanks for the tips - will check out those plug ins and have a fiddle. Yet more time to be spent infront of the computer - but oddly enough its kinda satisfying!

I’m a live engineer by trade - its been a long long time since I’ve done recording work. What do I do about overall volume levels on a finalised track - what’s the best way to decide? I can’t tell if my loudspeakers are too loud or too quiet and vica versa if my track level is too loud or too quiet - I don’t want to mix-down an mp3/wav that’s too loud or quiet (because that’s just plain annoying when you download one!)

I should have clarified - he is a beat boxer, but in the collaboration sense he’s more of a vocal loop artist. Its very, uh experimental shall we say. None the less I’ll have a fiddle with those plug ins and let you know how it goes :wink: . Cheers!

Trebor, is this the effect that thay seem to be using on the songs in the “Glee” TV programme? I’m just curious.


I think you’ll find there might be more than 1 effect that they use in Glee recordings… :smiley:

But yes extreme auto-tune software such as this is ever increasingly being used. I’ve seen it in TV commercials now - and its all over Youtube (look for the remixes of Double Rainbow…). Quite funny really, I really like how people can take the piss out of something with it - rather than trying to create plastic pop vocals. I’ll stop myself before I start blowing smoke out of my ears~

Yes but the “Cher” effect is with autotune with extreme settings: it wasn’t designed to be used in that way: you’re not supposed to hear it working, (correcting pitch).

PBS item on autotune …