Clicks Appearing When Pasting Ambient Room Sound

I think I got it.

Your pasted noise sample isn’t clean.

The second click in your posted example starts out “grass” or pure noise (blue line) but then rises into a real sound (red line). At that exact instant you cut back to the performance which gives you a click or pop (green line).

This effect will change depending on how long the pasted noise is. That gives you the magic of sometimes it pops and sometimes it doesn’t.

The extra sound could be anything. You moved a little in your seat or started to take a breath while you recorded your noise sample.

I think if you re-recorded your background noise and are more careful about extra noises, all these problems will vanish without special tools or masking.

If you can hold your breath and freeze for 6 seconds, that should cover the 5 seconds that ACX suggests for the end of chapter room tone. Everything else should be shorter than that. If you need longer, paste twice.

Do you correct the breaths before or after mastering? Some microphone systems create their own noise while recording and it doesn’t matter what you do.


Really? I was thinking that ANY noise that isn’t silence would get noticed, when the human QC happened, you see. Perhaps I’ve been very(too?) fastidious when it comes to extraneous noise. I use the multi view to see both waveform and spectrograph. I see them on the speccy.

But that seems to make it much easier than I have been thinking it is! I mean when ACXed, those bumps sound like, well, louder bumps, but not like a crack or pop that I would call distracting. It’s just I can hear them and I then think other people, customers and QC will, you see?

That explains that then. I notice too that you’re looking at the audio in the db view, not waveform, I’m not doing that; I think I might be missing out on information.

I’m using a Rode NT1 kit with AI-1 interface I’m in a treated area but I am keeping the record level low, when not speaking level rarely gets above -60 on PCM monitoring as i read -57/60 is a good quiet value somewhere when I was getting set up, getting peaks at -20, -18 and bumping up +2db to edit if it seems quiet. I edit breaths as I go, before ACXing.

My ideal workflow is minimal manipulation of files as possible. Ideally I just want to record, edit breaths and timing and ACX it and that’s that. Maybe a NR pass but the mic and interface are quiet enough most times. And then I don’t know if it’s best to NR the fles before or after ACX chaining.

I think I may be concentrating on the DETAIL of the file too much. It’s not uncommon for me to fixate.

Thank you.

I would try an expander plugin (like Couture) first to squish down the noise-floor (~6dB) when you’re not speaking, before resorting to Noise-Reduction.

The free version of Couture is sufficient, (other expander plug-ins are available).

That does look fancy! I’ve installed it and I’ll give that a shot, I’ve clicked at some things but I really am only guessing that it’s something to do with the knob bottom left, the other controls just make me sound funny.

I assume the red track is the modified waveform vs. the original? I mean I meddled with the bottom left knob, dropping it to -6db and got this on the waveform, the silence seems lower and the speech looks as it did, so that’s good, isn’t it?

Screen Shot 2020-06-16 at 21.53.25.PNG
But I mean I get good numbers on ACX check anyway.
Screen Shot 2020-06-16 at 20.44.31.PNG
But it’s always good to get better.

…and that’s because noise reduction is destructive, whereas this just - as you say - squashes the quiet bits to be quieter? Still new to most of these principles, want to get my head around them. It looks like sitting down and flapping the gums isn’t enough.

Turning all the knobs to three o’clock & “x1” is what I usually have.
That’s enough to knock the noise-floor down by 6/7/8dB,
but not so much that it sounds gated.

You mentioned that the noise floor was close to the ACX requirement of -60dB,
which is why I suggested Couture to get it down another ~6dB.

That’s the idea, that the speech is largely unaffected, (there may be a bonus of slight reverb reduction)

alright then, I’ll mess about with the settings and see what I can do. Thank you for the steer. Looking at the original pic posted of your plugin, you have it set like that there! I should pay more attention to what I see.

of course; as that’s the highest noise loor value (as it were) that ACX accept. So lowering it is good. I get that now. But also lowering record level would do the same thing? No… it means the speech would be lower too. So where I record low and give a little +2db bump up, I can use Couture to push the quiet down further, so I could have record level a little higher to start, if I understand this right.

I’ll delve deeper into this, but the take aways are cleaner slience for patch pasting and lower the noise floor further with an expander instead of NR. And perhaps don’t be so fixated on every spike on the spectrograph.

that being said, I could also bump up low(er) level recordings and THEN squash the floor down too, couldn’t i?

I would get the RMS to the required level, (using RMS normalize),
then apply Couture if I needed to bring the noise-floor down, and/or, dry-up slight room-reverb.

so squeeze it in before limiting on the ACX chain, if needed. Thank you, I’ll have a play.

Just to throw to another thread before this one gets put to bed, I’ve found that Couture seems to crash Audacity, at leaast my install does. Not all the time, but some of the time. And I was poking around the boards investigating expanders and I happened across this in the Nyqust dev board about Chris’s compressor. Would this do the same thing as Couture if set to negative values then? What’s the difference between the two? Or perhaps, is an expander a backwards compressor?

Exactly: full names are dynamic-range-compression & dynamic-range-expansion.

Steve made a Nyquist Plugin called dynamic mirror which can expand,
but it’s not as good as Couture: it doesn’t allow different attack & release times,
& I think there’s a limit on how much it can process in one pass, (an hour ?).

But it’s always good to get better.

No. It’s not. Audiobook Readers are a business and the business of being a business is to produce a marketable product for the least effort/time/materials. ACX is not going to give you extra points for lining up the most effects/patches/filters/corrections. There was one forum poster who claimed he was going to correct his audiobook reading word by word. He’ll be reading that book when he’s 92. No publication date yet, and his voice changed enough in the production time that the beginning and ending chapters no longer match. He’ll have to start over.

I produced an ACX Audition Performance using this …

… moved to my quiet bedroom. I cut it and mastered it. It passes Audacity technical testing, ACX technical testing, and it failed because I can’t read. Change the performer and the problem goes away.


That’s sage advice; it IS very easy to get caught up. I only started doing it for something to do…

I’m only doing thirty-minute bits tops. I’ve installed this as an alternative. Although truth be told, as with the Couture front end, I’m just sliding sliders and hoping for the best. I’m finding keeping numbers low seems to yeld the best results when set to mirror. But I’m just feeling my way really. Is there a suggested config for using this plugin as an expander?

Dynamic Mirror as expander suggestion.png
Not as good as Couture.

VST plugins can work more reliably in Audcaity’s competitor OCENaudio.

Thank you for showing me that. That resolution value is higher than I was using; I had 0.01, just (I think, if I understand it right) to squash the floor a little bit. It sounds nice at either really, without any need for NR, and that’s what I want to avoid. There’s not much extra noise at all rally, it’s jst for a polish.

Crikey, there’s so many out there! I might have a look at that, but I really quite like Audacity. Used it on and off for years, little reason to change.

Thanks again, all.

Ideally there should be two different time values on an expander: attack (~0.01) & decay (~0.10).
Steve’s dynamic-mirror only has one value for both, so a compromise is required, (IMO ~0.035 for speech).

Plug-ins specifically designed to expand are capable of a better job,
there are other free ones (other than Couture).
The older VST/VST2 plugins are more likely to work reliably in Audacity.

…so what do these values change? I realise we’re using a compressor as an expander so no attack/delay but what IS the control range then? If that’s the right way to ask the question.

“Control range max” is the threshold: the (RMS) level when the effect kicks-in, (in linear, not dB units).
When the waveform is above that (RMS) level, dynamic-mirror has no effect.
When the waveform is below that (RMS) value, it squishes it down further, (in mirror mode).

“Control range min” should be zero, unless you want to prevent the possibility of it squishing to flat-line zero.
(which would be a giveaway-sign that the audio has been processed).

“Control range min” must always be lower than “Control range max”.