De Essing on Audacity

Works quite well. It’s a little buggy on 2.1.2 and Mavericks.

When you disable it with the checkbox to listen back to the original, and then re-enable it, it seems to miss one click. The “playback” button stays highlighted and the playback doesn’t start. If you click on any other button, playback starts again and when you stop playback, it executes the previously clicked button.

I think it’s the plugin itself, as it doesn’t work at all, bridged in Reaper (it’s a 32 bitter). There’s no preview, nor an enable button…

Easily worked around. Now to Paul’s de-esser.

Back out to the 10,000 foot view briefly, once you sample a vocal crash and get it to sound right, you can select the whole performance and it will go down the whole thing and suppress all similar crashes right? What’s the longest performance it will correct?


The default-setting for the upper-frequency on Paul-L’s De-Esser is 8kHz, which is way too low.
Most people can hear up to ~14kHz , 15kHz seems like a better value for an upper-limit …

So the default settings make the effect slow, from what you said about using more than 6 bands.

How much more effective is the plugin at 10 bands than 3 bands?


The real problem with default settings is the upper-limit set at 8kHz, that means a large chunk of the sibilance is untouched by the De-Esser. Novices who try Paul-L De-Esser on the default settings ,(and don’t know how to adjust them), will declare “it don’t work”.

Wikipedia have underestimated the upper limit too …
female voice sibilance [Dawn Maxfield].png

Found it.

This is a posting from May 2015 before a lot of our tools became available. Over the course of several weeks, all the problems got resolved and I believe the poster is cranking out book after book.

It had an Essing problem which I cured with simple equalization.

Attach 1: Overly Crisp Clip.
Attach 2: Custom Equalization curve applied.
Attach 3: Actual xml curve.
Attach 4: Graphic of the filter.

All I did was run Analyze > Plot Spectrum and made note of the two high frequency haystacks clearly in the wrong place. In addition to being easier on the ears, the shoot-to-the-moon sharp peaks in the blue waves vanished making it a lot easier to hit ACX and all without making the voice muffled.

I tried going back with the original clip to the DeEsser tool and eventually got it to suppress one (1) overly sharp Ess sound.

Screen Shot 2016-02-22 at 20.49.05.png
CrispPatch2.xml (664 Bytes)

Using my suggested defaults for Paul-L’s De-Esser , processing time is ~1/20th the playback time , ( on my 5+ year old computer ) . Using your “CrispPatch2” equalization-curve , processing time is ~1/30th the playback time, so it is faster, but IMO the De-Esser does a better job at reducing excessive sibilance …
''Driver's Seat'' comparisons.png

Of course, the equalizer never claimed to Solve the World’s Problems. Just the worst ones and it uses easily discoverable targeted frequencies. I’ll try that solution when I get back.

It occurred to me the level or slice control is actually answering a question nobody asked. It’s not at what level should I start working, it’s at what level compared to everything else. It does nobody any good to trigger the correction at -20 when in a particular instance, no part of the performance rises to -20.

And that brings us to the conundrum of which corrections do I apply first?


I agree : in an ideal-world the De-essing threshold should be dynamic : in-part dependent on the recent loudness , (i.e. “ride the envelope”).