Removing Pop spikes

[tlm]

I've tried many things to remove all 'pop' spikes in audio. Recently, I've come close using the Label Sounds plug-in I found here in this forum: viewtopic.php?t=110171 Once the spikes are labeled, I simply remove them by selecting the audio and label tracks and using Edit> Labeled Audio>Split Cut.

The result is okay, but my question is: how can I improve?

The attached audio and image demonstrates. The spikes shown as 1-8 in this image are what I want to remove, and you can see that I find most of them using these settings of the Label Sounds plug-in:
Threshold level: -17dB (or -18dB), Threshold measurement: RMS level, Minimum silence duration: 0.01s, Minimum sound duration: 0.01s, Label Type: Region labels, Maximum leading silence: 0s, Maximum trailing silence: 0.02s, Maximum number of labels: 9999, Label text: ##1

Notice that going from -17dB to -18dB allows spike 4 to be labeled at the cost of labeling non-spikes, and even at -18dB spikes 7 and 8 are not found.

pop.jpg (71.71 KiB) Viewed 783 times

Thanks for any help. I'd also be happy to receive other advice on my method.

Other info: I'm using Audacity 2.4.2 on Mac OS 10.15.7 and I zipped the attached file using Keka to stay within the 2MB limit (it opens fine with the standard Mac unarchiver)

[steve]

There's a plug-in for muting pops: viewtopic.php?p=122958#p122958

[Trebor]

If there were only a few pops, reducing them individually with Audacity's spectral editing tools is feasible ...

reduce pops with Audcaity's spectral edit tools.gif (155.43 KiB) Viewed 777 times

before-after.wav

(461.46 KiB) Downloaded 26 times

As there are so many, I'd be tempted to re-record with a pop-filter.

[tlm]

There's a plug-in for muting pops: viewtopic.php?p=122958#p122958

This is great, thanks Steve.

But I found the major limitation being missing the smaller smaller spikes verses distorting the rest of the audio. It seems the tool detects spikes based on a dB threshold, but it seems to me that since pop spikes are quite narrow there should be a way to detect spikes based on their sharpness.

In scientific spectroscopy, the peak width at half height is used to distinguish sharp verses broad peaks. In the image below I try to make my analogy clear by showing a crude drawing of the 'envelope' defined by an Audacity sound wave.

envelope.jpg (94.58 KiB) Viewed 765 times

I don't know how to make plug-ins, so I don't know how hard that would be. But, my idea is: (1) define a waveform envelope for the audio based on user defined resolution, (2) apply the pop mute based on a peak width at half-height for that envelope. Step 2 means finding a local maximum, going to half that height and measuring the width -- which for the narrow peaks should be easy. It could be that dB threshold could be entirely replaced by the width at half-height, or a combination of might be needed.

[steve]

It seems the tool detects spikes based on a dB threshold

That's correct. Anything over the "threshold" level is reduced.

but it seems to me that since pop spikes are quite narrow there should be a way to detect spikes based on their sharpness.

One of the use cases that this effect was written for, was to reduce the level of applause in a recording of a (classical) music recital. In this case the "bad" audio was not a "spike" but rather a sustained high level of noise.

The best "solution" for the pops in your recording would have been to avoid the pops in the first place.
It sounds as if the pops could have been caused by blowing on the microphone. If that's the case, then a mechanical "pop filter" would be the best solution.

[tlm]

Steve, Thanks again for your comments. I'm well aware that doing the audio right in the first place is obviously the best way to deal with problems like this. Believe me I try to help those involved in recording to do just that. But I know from years of experience that's just not gonna happen for the audio I deal with since it comes from many different people in many different environments.

Given that reality, could you address the points that I make? Or if not hopefully someone else can. Thanks.

[steve]

In the waveform (above), the red line tracks the "amplitude envelope" of the waveform.
The "Pop Mute" plug-in creates an envelope follower signal like that, and then applies negative gain when the envelope (red line) exceeds the threshold level.

Think what would happen if a very low frequency High Pass filter was applied to the envelope - higher frequencies would pass through, and lower frequencies would be attenuated. In theory at least, this would allow you to set the threshold to a level that catches the "higher frequency" narrow spikes while missing the "low frequency" gradual rises and falls.

[Trebor]

... Given that reality, could you address the points that I make? Or if not hopefully someone else can. Thanks.

Paul-L's DeClicker plug-in can be configured to DeThud,
(as clicks & plosive-thuds are similar phenomena on different timescales)



NB: Without human-supervision, (or AI),
DeClickers will reduce/remove clicky parts of speech, as they don't speak the language.

DeBreath plugin$ are worth a look, (as plosives are breath, identifiable by lack of harmonic content).

[waxcylinder]

You might want to try ClickRepair - it costs a little (you do get a 140day free trial) - it worked magical wonders on my LP transcriptions.

See this sticky thread: viewtopic.php?f=28&t=1994

WC

[rcarlberg]

In my experience (>20 yr) cleaning up LPs, none of the automated declickers is worth a crap. They all leave artifacts.

The only process that results in a clean file is manually declicking, zooming in to the peaks and using the "repair" tool to flatten them. Even then you have to be careful -- if your area of damage is more than 128 samples, multiple runs at declicking can result in aliasing.

Sadly, there's no alternative to hard work.