Noise Floor

I’m a newbie here, so hello! I’ve had a lovely time over the last few days looking through past messages and have found most of the answers to my questions without having to ask them. One thing I can’t find though…

I’ve got an Apogee Mic that only works with my phone or iPad, not my computer. That’s okay, it means I can avoid a whole lot of noise issues - no computer fan etc. I have been recording on the iPad in the Twisted Wave app, then dropboxing it to my computer in wav format and editing it in Audacity. My question relates to the fact that when I select a moment of room tone in Twisted Wave, it tells me that the noise floor is about -53dB, but when I analzye it in Audacity, it looks much quieter than that - closer to -65.3dB.

I’ve attached two files 0 the same snippet with a bit of room tone at the start (I edited out the breath just before speaking). Running the ACX check on that, it tells me that the NoiseFloor is -65.8dB, but that it fails to meet RMS levels. (This is the version that, on Twisted Wave on the iPad registers about -53dB.)

The second file is the same snippet, normalized to -3.2dB in Audacity. This version does meet ACX requirements. The NoiseFloor is now -62.4dB.

So the question is, how accurate are both of the two systems? It’s a bit alarming to get such a difference. (I can further lower the noise floor by turning off more appliances, but this version is with the main offenders turned off - computer, fridge, fish tank…).

Thank you for your time!

They can all be perfectly accurate, but you do have to define your terms.

We designed ACX-Test as an add-on to Audacity in order to simulate the technical specification robot you’re going to hit when you submit to ACX.

The tool appears under Analyze and the first three values are hero. The tool will tell you if you passed and by how much.

Noise can be measured a number of different ways. Like it says in the brief, it’s what happens when you stop talking. You can measure the peaks, that is, watch the sound meters (assuming you have them set for good enough sensitivity) and get a rough reading from that. That may be what the other program is doing. That’s easy to do, but not particularly useful.

You can measure the RMS value of the noise. Without knowing what that is, it’s the way both ACX-Check and ACX submission work. It’s the amount of energy in the noise. As a rule you can’t get there by looking. You have to have a tool measure it for you. Audacity has a manual tool called Analyze > Contrast which will tell you many of these values, but it’s much more difficult to use. ACX-Check is just so much easier.

Once you get the energy or RMS value of the noise, you don’t have to stop there. There are variations on noise measurement. There is a legal one which applies the “A” curve to the measurement. You can have a thunderstorm and bats flying around the house (work with me here) and some noise instruments will tell you there is a terrifically high noise value—and you can’t hear a thing. That’s because the bats are too high and the thunderstorm is too low. The “A” curve ignores any noise humans can’t hear. That’s the one hazardous noise environment measurements use.

But ACX doesn’t use that one. They use the one that can hear bats and metro-busses, so it’s good to know that. We have a trick you can use if you almost but not quite pass noise and could use a little boost. We have a metrobus filter.


You should know that twice now I was able to set up a simple microphone arrangement in my very quiet third bedroom and produce an ACX Compliant voice clip by simple adjustment of the volume. So no, laundry lists of corrections, adjustments, filters, effects and manual patching are not needed for a generic ACX posting if you’re doing a reasonable job with good equipment. I will admit I probably could not read a book like that, but I just wanted proof of concept.

So that’s what you’re doing.

I designed a forum voice test format.

This gets everything we need into one pass and posting. We can usually tell you how to get to ACX submission from where you are and how to tune the corrections if needed. Please note we can’t take corrections out of a clip, so submitting a fully corrected clip and asking, “What’s wrong with it,” doesn’t go anywhere.


I analyzed the raw clip and found it a little weak.

– Select the whole clip or show by clicking just above MUTE.
– Effect > Normalize: [X]Remove DC, [X]Normalize to -3.2 > OK

And that gave me the attached.
Screen Shot 2016-06-12 at 22.04.55.png
So you’re good to go. That’s very similar to what I did to make my clips pass. In English, make it slightly louder.

You may decide that the clip is too bright, crisp and forward. I don’t think it’s objectionable, other elves may have thoughts. That’s the only thing I found “wrong.” Your clients may love it.


1 Like

By the way you have my affliction. Your voice is noticeably non-symmetrical. Your down blue waves are longer than your up ones.
Screen Shot 2016-06-12 at 22.14.27.png
That can create compliance problems, but your voice seemed to pass OK.


I’ve never noticed the asymmetry!

So, it looks ok, then. I have an idea that my mic is a bit treble-y and have been toying with adding a little something to compensate, but first things first. At present though I’m really just normalizing. I might have something from a professional studio and a fancy mic. Will see what I have…

Thank you so much! It was a little worrying. I will ignore the app and focus on the information from Audacity.

I really appreciate the help.

Oh, and you answered a question I was thinking but hadn’t yet asked! The noise floor (a) definition - thank you! So it sounds as though I can safely ignore that value.
Thnak you again

my mic is a bit treble-y

Crisp, yes. If you do that too much you fall over into “harsh.”

sounds as though I can safely ignore that value.

You have to pay attention to all of them, but yes, you do seem to have a quiet microphone and recording environment. This just kills newbies. Nobody can hear their environment sounds until they get captured in a sound clip. “I had no idea the refrigerator was that loud!”

-60dB is a pretty serious number. Most living rooms or offices can’t hit that and that is where most people set up. The first time you miss the peak/RMS/Noise conformance you start the dance of the effects. “If I make the RMS (volume) louder, the noise comes up. I can Use Noise Reduction but then my voice starts sounding funny…”

The family had a vacation boarding house in Upstate New York. Famously, at night, the environment sound would go to zero and nobody visiting from noisy New York City could sleep.

Just to dispel another mystery. Effect > Amplify and Effect > Normalize just turn the volume up and down. Full stop. That’s all they do. They do it with different controls and paying attention to different parts of the sound. “Normalize” sounds like something you should be using to make your show normal so it passes ACX. Nice try. It’s called that for historical reasons.


Miranda, we LOVE people like you that do the reading :sunglasses:

I assume youve founf the Manual with allis Tutorisls etc:

Note that the image of the Audacity window on the front page is actually an imagemap meaning that you can click on various parts of it to drill through to the appropriate page in the Manual.


I messed around turning off the most likely offenders one at a time until I got it to where I thought it needed to be. There are still things running, though, so I’m sure I could get it down a little lower. I image it’s a diminishing returns thing, though, and it sounds as though my set up is good enough for the moment.

I think the mic/iPad combo helps a lot. My computer is the noisiest thing in that room.


I didn’t know it was clickable, I’ll go check.

So much amazing information here, not to mention all the plugins.

I designed a first-pass de-harsher, but I need to wait until I can get on a different machine to present it to you.


I’d love to try it. No rush - it’s bedtime in the land of Aus!

Attached AbegaleFlorist.xml. This is a way to add equalizer (tone controls) to specific parts of the sound. Download and put it on your desktop or somewhere you know where it is.

Import or open the raw sound clip. Do the Normalize thing.

– Select the whole clip or show by clicking just above MUTE.
– Effect > Normalize: [X]Remove DC, [X]Normalize to -3.2 > OK

Effect > Equalization.
Save/Manage curves > Import. Point it at AbegaleFlorist.xml > Open. It should appear in the Curves list. > OK.

You would think it would automatically pop up ready for use, wouldn’t you?


While Equalizer is open, select AbegaleFlorist from the list of pre-baked curves.
Screen Shot 2016-06-13 at 6.18.22.png
And OK to apply that filter to the work. If you pay attention, you can see tiny parts of the blue waves disappear from the timeline. If I did it right, those are the sharp sibilants. You can use Edit UNDO and Edit REDO to flip back and forth between filter and not. See if you like it. It’s entirely customizable. If the effect is too dull (I don’t think so), I can make a half-way curve.

Enjoy (just getting up).

AbegaleFlorist.xml (297 Bytes)

I know this gets lost in the weeds, but you have a terrific voice.


There’s actually a piece of this I left out. I usually include a picture of what the “curve” looks like so you know you’re close. Attached what Audacity Effect > Equalization looks like when you apply AbegaleFlorist.xml. You may need to scan around and change the browser size a bit to see the whole thing.

On the right-hand side there around 10,000, that’s where escaping air, tinkly bells, crinkly newspaper, and harsh “SS” sounds in speech live. So that’s a fancy-pants way to turn the treble down a little. I didn’t make that up cold. Audacity has tools to tell me where the different parts of your voice live, so I do have some guidance.

Screen Shot 2016-06-13 at 11.55.00 AM.png

Thank you! I can indeed see the little blue bits vanish. I didn’t do any compression - is that something you’d typically recommend?

I attach a snippet of an ad I did at a radio station. It came to me as an MP3 and I’ve snipped it in Audacity and exported it as a wav. I see that it would fail ACX! Noise floor is much too high, and the RMS is out too! I think my voice sounds a bit “fuller” on their recording than on mine, but I’m fascinated to see that my home office is (seems to be?) quieter than their booth! I also hear that I was too close to the mic - I’m popping a bit. :frowning: That’s hard for me to monitor when I’m doing stuff there as I’m in a booth and don’t get to see the software. (I hope I’m allowed to post this, it’s a snippet and it’s already gone to air.)

Thank you again!

I hope I’m allowed to post this,

It generally gets our attention when someone is blatantly trying to sell us something (as recently happened on the forum). Sample clips are perfectly fair game. Plus, if a forum elf tells you to post it, you’re covered.

That’s hard for me to monitor when I’m doing stuff there as I’m in a booth and don’t get to see the software.

They’re supposed to be watching levels, but you should hear what you’re doing in the headphones, right?

If it’s pure broadcast, they can get away with a lot by covering up the vocal with music and other production. The instant they go on record as recording for hire for audiobooks, the jig is up and they have to conform just like everybody else.

I think my voice sounds a bit “fuller” on their recording than on mine

That’s proximity effect and it happens when you get closer and closer to a directional microphone. For guys, it gives us the robust broadcast voice. For women, it gives a more mature, smooth voice.

Until you get too close. Then it starts popping. That’s what the headphones do. Popping should never have made it past the first sentence. “What’s that funny thumping sound when I say my “P” words?”

If you don’t know where the clip came from, it’s possible it’s from an air check and it’s been through transmitter processing to air and back. You can ask them.

I didn’t do any compression

Works for me. It’s possible if you read for a very long time, that an overall chapter may fall out of compliance because you’re wandering in volume. In that case, you may decide to add some gentle compression to lower the hot bits and raise the quiet ones.

But maybe not. This is an excerpt from an ACX training video about mastering.

I’m pulling your clip down now.


The clip passes very nicely. Yes, it’s been through some compression processing so the RMS (loudness) is actually higher than ACX would like it. All I did was reduce the overall volume (Effect > Amplify: -2 > OK.) and both Peak and RMS fell straight into compliance.

What about noise? The noise measuring tool needs a certain amount of quiet to measure and you don’t have enough. I measured the noise The Old Fashioned Way with the manual tools and it’s something like -80. So yes, you did manage to fake out ACX-Check, but there’s nothing wrong with the clip.

Oddly, I don’t hear any popping or other distortions other than the voice being slightly crisp.


How to measure noise the old fashioned way.

Drag-select the silent portion. I used the portion at the beginning .05 to .25.

Analyze > Contrast > Measure selection (top one) > Read the value > Close.

For this one clip for that one value, you have to do it the old way. I think ACX-Check needs a half-second of silence to work. I can check that.


Yup. 1/2 second. There is no half-second silent stretch in the clip, so the tool went nuts.