New vs. Old Noise reduction settings.

Hi all! Just upgraded a couple of our machines and playing around with the new version of Audacity.

Coming off of using 2.0.4 there’s a couple settings that have changed drastically, and I was hoping for a little assistance with the Noise Reduction tool.

Before I would use Reduction 13, Sensitivity 0, Smoothing 150Hz and an attack of 0.15 (In other words, the defaults, except for the reduction)

It was almost PERFECT for removing the noise floor of our studio setup without much altering or fiddling, and without affecting the audio quality any.


The defaults on the new Audacity seem WAY less than cooperative. I’ve got the new settings at 13, 12, 2 and it works, but it’s not as good, and since the way in which the values are computed/presented have changed I do not have a good feel for them yet.

Anyone got a magic combo that’ll pretty much mimic the effect of the old, fine-tuned reduction/removal tool? (I could sit here and fiddle for an hour but I’m guessing someone - probably Steve - has already figured out a way to do it, and why reinvent the wheel if I can just ask? ; )

P.S. I’m aware that noise reducing should be evaluated per audio file, it just happened that the defaults more or less were perfect for this sound booth, as well as I understood the settings on the old one for other sound sources as well, and could adjust accordingly. This one I don’t, and so can’t, and am hoping for a good baseline from which to begin playing with it.

Mostly I leave the second two controls at their defaults:
Sensitivity: 6
Frequency Smoothing: 3

Then adjust the first slider, Noise Reduction (dB), to give the required amount of noise reduction. Too high gives a bubbly metallic effect - too low and there’s not enough noise reduction. For a high quality recording (low background noise), somewhere around “6” is usually enough. For noisy recordings, higher will probably sound better, but may start to sound a bit metallic / bubbly.

A followup question if I may.

Before the Frequency Smoothing was measured in Hz (which I understood.)
Now it has a “bands” number, and I’m not quite sure how this relates. Is it a ratio/percentage of your overall sample rate?

I wouldn’t mind notes on that, either.

I’ve been recommending 6, 6, 6 for good, graceful reduction on voice recordings and that has proven effective. I’ve been assuming x, 6, 3 is more appropriate for music and more complex recordings. I’ve never checked it.

It’s not unusual for some processes to take shortcuts so as to get the work out the door quickly. What are the settings if you want maximum possible quality…I’ll wait?


It was almost PERFECT for removing the noise floor of our studio setup without much altering or fiddling, and without affecting the audio quality any.

There were others that claimed very desirable results from Noise Removal. I think y’all are unicorns. I was never able to get the tool to do anything valuable, ever. I think there were many in my camp which is why the tool was changed. That and too many people wanted it to actually Remove Noise, which it would not do. So it was lying right at the label on the tin.

I had a joke about one of the settings was perfect if I could figure out how to set it between pixels on the screen. The existing range was too little and too much—no option.

So that’s what happened. You’re in an exclusive group.

I suppose this is a bad time to suggest better studio noise abatement if you routinely need that stiff a post production reduction?


I suspect that the “Hz” was based on assumptions about the sample rate, rather than actually being “Hz”.

The number of bands is similar to the old Hz setting.
Although not really accurate, you can think of the noise reduction as a series of notch filters that filter out specific frequencies of noise. The greater the number of bands, the wider the notches are, so they tend to cut out more noise, with less metallic bubbling, but removing more of the sound that you want to keep.

If I recall correctly, there are 1024 frequency bands, so at a sample rate of 44100 Hz, the total frequency bandwidth is 22050 Hz, and each frequency band is 22050/1024 = 21.5 Hz.

Noise Removal was famously useless in the face of white/pink/brown noise. Yet another reason it was abandoned. What’s the difference between the two systems that makes Noise Reduction actually valuable to reduce “microphone preamp hiss?”

I think the explanation may be a little too fuzzy because strenuous application of Noise Reduction will cause sibilance and harsh mouth noises to go up, not down.


THAT makes more sense. And now that I play with the bands setting I routinely get good results off the new tool now.

As to the suggestion regarding how stiff our removal setting was, it was for one particular mic on the array that we keep the gain dialed WAY down, so that we could catch loud yells and not clip.
However, when you’ve got it at -20dB to the rest of the mics, and you bring it up to volume, you get just enough of the audio out of it to salvage the take, but it requires a lot of compression and - yup! you guessed it - noise removal, because you’re cranking the noise floor up to the stratosphere.

It’s a spot-correction tool, not an entire track at a time, in these cases.

EDIT: I should probably specify we do mostly character voices for games. Lots of screaming, and the aim is short, standalone, unmixed vocal clips. Hence the “unicorn” status you’ve bestowed and I’ll gladly acknowledge and accept. ; )

so that we could catch loud yells and not clip.

I’m fascinated anyone thought to do that. Works for me.

It’s a spot-correction tool, not an entire track at a time

I can well imagine correction for integration into the show would be serious, but worlds better than the splattering and clipping of a microphone and preamp overloading.

now that I play with the bands setting

And behold, the documentation problem. You can’t put: “Mess with it until it works” into a wiki. It’s not expandable. If you have two tools like that, post production is stunningly difficult and if you have three, you will never get your show out the door. The interactions have interactions.

It’s a math thing. Three factorial? I try not to think about that on an empty stomach.

This is where we corral the Noise Reduction Developer.

‘What were you thinking when you wrote this (past killing Noise Removal)?’


so that we could catch loud yells and not clip.

I’m fascinated anyone thought to do that. Works for me.

Yeah, it’s INVALUABLE if everyone is getting excited and into character. Games have a LOT of scenes where you shift from normal volume to outright banshee screeches, and if you have a good high SPL mic and your gain set properly it’s not USUALLY a problem, but SOMEtimes your enthusiasm gets the better of you. XD

Actually, for voice-only, especially the really LOUD stuff that we use this for? It’s not bad. We’ve got an EQ profile we already tweaked to perfection to accommodate the slight difference in the pickup pattern when you pad the mic that heavily, the compression tool also helps richen it up a bit, and the mic itself matches the others, so when you’re just cutting out one line, and one that’s screamed anyway, the worst you have to do, if it’s a REALLY difficult one is splice the lead-in or tail of the line from the main track carefully so there’s no noise artifact. (Otherwise the lead in/trail off is quiet and odd sounding.

You get good at it, and you’re looking at maybe an 8-12 second operation. 20-40 in REALLY difficult circumstances. ; )

And it’s also better than calling people back in, or having to blow out your own vocal cords over and over. = )

Yes. Nobody is expecting screamed dialog to sound like anything anyway, so you win.

It’s not often somebody comes up with a trick I haven’t heard of.

[Writing that down]


Thanks to the thread. Was able to get some work done after watching a training session and not finding “Noise Removal”.