invert phase, and sidechained gating—in rack w/stock devices?

[guitfnky]

what it says on the tin...are there any stock devices that will allow me to:
-invert phase (want to avoid using the mixer for this, and want to stay native to Reason if possible)
-gate which will open based on a sidechain signal I feed into it (again, preferably using stock devices)

looking for simple solutions with one or maybe two devices at most for each of these, if possible.

[Timmy Crowne]

Thor can perform both the phase inversion and sidechain gating.

Cable your main audio signal to an input pair on the rear. Then in the modulation matrix, send the signals to an output pair with an amount of -100. This accomplishes the polarity inversion.

For gating, click the scale column for the modulation routing you just set and select Global Env as a scale factor. The amount parameter here is our gate range or depth. Find the Global Env module above the mod matrix and set it to full sustain, with a release around 100ms for now. Now cable your sidechain audio signal to a 3rd input on Thor's rear. Create another routing in the mod matrix that sends the sidechain signal to Global Env Gate. The amount parameter here is our gate threshold.

Finally, play with the threshold and Global Env parameters to taste. Hope this helps!

[guitfnky]

Thor can perform both the phase inversion and sidechain gating.

Cable your main audio signal to an input pair on the rear. Then in the modulation matrix, send the signals to an output pair with an amount of -100. This accomplishes the polarity inversion.

For gating, click the scale column for the modulation routing you just set and select Global Env as a scale factor. The amount parameter here is our gate range or depth. Find the Global Env module above the mod matrix and set it to full sustain, with a release around 100ms for now. Now cable your sidechain audio signal to a 3rd input on Thor's rear. Create another routing in the mod matrix that sends the sidechain signal to Global Env Gate. The amount parameter here is our gate threshold.

Finally, play with the threshold and Global Env parameters to taste. Hope this helps!

I think this will! I’m only moderately familiar with Thor’s matrix, but that seems to make sense and I think I’ll be able to follow it easily enough when I get a chance. if I can get both of these done with a single Thor, I may only need one additional device, which would be pretty amazing for what I’m trying to do. if it works, I’ll share it later.

many thanks!!!

[Timmy Crowne]

Happy to help. Sounds good, please do share. I love to see techie stuff like this in Reason!

[selig]

I checked again just to be sure, and it oddly appears Thor is the only native device in Reason besides the SSL Channel that has a way of inverting polarity.
As for a gate, you have the excellent SSL Gate now available outside of the mixer in the Channel Dynamics device, which IMO works better than using Thor as a gate. If you run a sine through Thor using it as a gate, you'll see a lot of intermodulation distortion when using a release time faster than around 500 ms. At 100 ms there's all sorts of side bands being generated, while at 100 ms release and fast attack the SSL gate is still totally clean. I believe this is possibly due to using a bipolar audio signal as a trigger. A compressor/gate will first process the audio input with an envelope follower, which then produces a unipolar signal better suited to control a VCA for gating/compression.
[EDIT: the Release shape on the SSL Gate is also better suited for audio than Thor's envelope IMO, since it's more exponential.]

[guitfnky]

I checked again just to be sure, and it oddly appears Thor is the only native device in Reason besides the SSL Channel that has a way of inverting polarity.
As for a gate, you have the excellent SSL Gate now available outside of the mixer in the Channel Dynamics device, which IMO works better than using Thor as a gate. If you run a sine through Thor using it as a gate, you'll see a lot of intermodulation distortion when using a release time faster than around 500 ms. At 100 ms there's all sorts of side bands being generated, while at 100 ms release and fast attack the SSL gate is still totally clean. I believe this is possibly due to using a bipolar audio signal as a trigger. A compressor/gate will first process the audio input with an envelope follower, which then produces a unipolar signal better suited to control a VCA for gating/compression.
[EDIT: the Release shape on the SSL Gate is also better suited for audio than Thor's envelope IMO, since it's more exponential.]

ah, good to know about the Thor distortion—I have no idea how to “hear” such a thing, especially in my use case, so I’ll stick with the rack SSL gate (totally forgot you could sidechain that for some reason).

I was really surprised there wasn’t a phase invert on either of the rack mixers. I got excited when I remembered there’s an invert out on the CV splitter/merger, but alas, no equivalent on the audio one.

[selig]

I checked again just to be sure, and it oddly appears Thor is the only native device in Reason besides the SSL Channel that has a way of inverting polarity.
As for a gate, you have the excellent SSL Gate now available outside of the mixer in the Channel Dynamics device, which IMO works better than using Thor as a gate. If you run a sine through Thor using it as a gate, you'll see a lot of intermodulation distortion when using a release time faster than around 500 ms. At 100 ms there's all sorts of side bands being generated, while at 100 ms release and fast attack the SSL gate is still totally clean. I believe this is possibly due to using a bipolar audio signal as a trigger. A compressor/gate will first process the audio input with an envelope follower, which then produces a unipolar signal better suited to control a VCA for gating/compression.
[EDIT: the Release shape on the SSL Gate is also better suited for audio than Thor's envelope IMO, since it's more exponential.]

ah, good to know about the Thor distortion—I have no idea how to “hear” such a thing, especially in my use case, so I’ll stick with the rack SSL gate (totally forgot you could sidechain that for some reason).

I was really surprised there wasn’t a phase invert on either of the rack mixers. I got excited when I remembered there’s an invert out on the CV splitter/merger, but alas, no equivalent on the audio one.

As for hearing the Thor distortion, I wasn't sure what I was hearing so I used a sine wave and viewed it on the spectral display. It's super easy to see any distortion when using a sine wave in a spectrum display because it's any frequency that's preset other than the one the sine wave is generating! When you see additional sines above the original, it's probably saturation/clipping (harmonic distortion). When you see sines above and below the original it's either aliasing or intermodulation (inharmonic) distortion.

No simple polarity inversion is an odd omission IMO, and when building Combinators for sharing that need inversion Thor is the only option, though it sure seems like such overkill! For my own "in house" projects I of course use Selig Gain. There were 3-4 "missing" features in Reason that inspired that RE, one being the need for a simple polarity invert device. Other missing features were the inability to automate fader gain by decibels, no dual pan knobs for left/right (with optional pan laws), and no simple way to scale/invert CVs or to generate a static/constant CV signal. Thor can do all of those except automate by decibel, but again, such overkill!

[guitfnky]

alright, I've got it working pretty well...mostly.

I built an inverted phase de-esser. basically it works like this...

split the signal three ways
-first signal goes to a line mixer
-second signal into a Thor for inversion (thanks Timmy!)
-second/inverted signal then goes into a Channel Dynamics rack extension to be gated (thanks Selig!)
-third signal goes into a vocoder (EQ mode) to identify all of the eSSeS, and then into a Channel EQ for more HPF, all the way up to the max 4khz
-third signal/eSSeS signal gets split--one goes back into the line mixer so it can be listened to if necessary, but the important one gets sent to the sidechain input of the Channel Dynamics, which is controlling the inverted signal

it took a bit of work to dial in the gate settings, and the level I needed to set its output to in the line mixer (you can't leave it at the same setting as your clean--i.e. first--signal, or it will completely obliterate your eSSeS with this particular setup), but once I had it up and running, it actually sounds more transparent than most every other set-and-forget de-esser I've tried recently (and I've tried a lot). of course, I've been listening to my own voice and the sound of eSSeS for hours now, so my ears may just be shot.

anyhoo, the problem is, if I get to a phrase where I'm really pushing the vocal, that starts to hit the de-esser as well, so it's doing the usual thing any gate does when it's not triggering the way you want--causing problems.

I know that's sort of a thing with most every other de-esser, so I don't expect to eliminate the problem completely, but I'm looking for ideas/best practices on how I can minimize the impact of the non-eSSeS. how are pro de-esser plugins designed to handle that? does it take a lot of additional processing (what I'm getting at is, is it possible to do using stock devices).

any advice would be super welcome!

[Timmy Crowne]

That sounds pretty cool! As far as keeping the non-esses from triggering the gate threshold, I suspect you may be able to achieve it by patching a compressor on the sidechain signal *before* it hits the vocoder. I would try setting the threshold so that you only get gain reduction on those phrases where you're really pushing. I'm probably wrong, though haha

And Selig, that is very illuminating what you said about the sidebands and the need for an envelope follower. I just want to say I'm grateful for your expertise in these areas and your taking the time to share it with us.

[guitfnky]

That sounds pretty cool! As far as keeping the non-esses from triggering the gate threshold, I suspect you may be able to achieve it by patching a compressor on the sidechain signal *before* it hits the vocoder. I would try setting the threshold so that you only get gain reduction on those phrases where you're really pushing. I'm probably wrong, though haha

And Selig, that is very illuminating what you said about the sidebands and the need for an envelope follower. I just want to say I'm grateful for your expertise in these areas and your taking the time to share it with us.

that sounds like it could work! I may also do some more creative sidechaining so I’m triggering the compressor only when there’s an abundance of stuff below my sibilant range (there are some esses during the louder part), but that shouldn’t be too difficult. I’ll have to play around with it, but that might just do the trick—awesome suggestion!

[guitfnky]

that seems to have done the trick, thanks again for the suggestion, Timmy!

I need to do some final tweaking of the combi knob control settings to get them into good usable ranges, but it’s pretty damned solid, if I do say so myself.

once I’ve finished that up, I’ll share the patch, and maybe do a (hopefully quick) video walkthrough of how it all works. sounds REALLY nice, so far, at least to my ears. I’ve never been this excited about a de-esser before.

[selig]

That sounds pretty cool! As far as keeping the non-esses from triggering the gate threshold, I suspect you may be able to achieve it by patching a compressor on the sidechain signal *before* it hits the vocoder. I would try setting the threshold so that you only get gain reduction on those phrases where you're really pushing. I'm probably wrong, though haha

And Selig, that is very illuminating what you said about the sidebands and the need for an envelope follower. I just want to say I'm grateful for your expertise in these areas and your taking the time to share it with us.

Well I wouldn't have discovered it if I hadn't compared using Thor side by side with the SSL Gate, so thank YOU for suggesting the Thor gate trick. Thor's envelope would work fine if triggered by a clean gate signal, which can be derived from Pulverizer's envelope follower. There will be a few ms delay doing it this way, but the advantage is it will be a clean trigger.

[selig]

anyhoo, the problem is, if I get to a phrase where I'm really pushing the vocal, that starts to hit the de-esser as well, so it's doing the usual thing any gate does when it's not triggering the way you want--causing problems.

I know that's sort of a thing with most every other de-esser, so I don't expect to eliminate the problem completely, but I'm looking for ideas/best practices on how I can minimize the impact of the non-eSSeS. how are pro de-esser plugins designed to handle that? does it take a lot of additional processing (what I'm getting at is, is it possible to do using stock devices).

any advice would be super welcome!

That's the problem with any dynamics based de-esser - sibilance will be reduced based on how loud it is rather than whether or not it's present. You would think this would be want you want, and much of the time it works. But if you have a section of a song where the vocal gets louder/quieter overall, the de-essing won't work well because it's reduction is based on it's overall level rather than relative to the main voice. Meaning, if you set the threshold to de-ess the loudest sibilance, it will reduce the sibilance relative to the main voice by a certain amount. But then if the voice gets soft, the sibilance will ALSO be soft and thus will fall below the threshold. So in that part of the song the sibilance won't be reduced at all, and the voice/sibilance relationship will remain unaffected.

Most of the time a dynamics based de-esser is built by simply feeding the EQ'ed (isolated) sibilance frequency to the side chain of a compressor/limiter. Alternatively, a crossover/splitter can be used to process just the sibilance range, ducking it when energy exceeds the threshold in the sibilance range. Another issue with this approach is that sibilance doesn't always occur at the same frequency, even over the course of one song with one voice (let alone different signers on different songs). So you have to either feed a wide band of frequencies to the detector or automate a narrow band to follow any changes. This approach can also miss other constant sounds that may need reducing such as F, T, and Ch sounds.

What exactly do you mean by "eliminate the impact of the non-esses? What do you hear happening to the main voice (non-ess) currently?

Shameless self promotion mode: ON!
Of course I HAVE to mention the Selig DeEsser here, which was developed precisely to work with situations like this. As for transparency, it doesn't do ANYTHING to the audio path except for a level adjustment - no filters are in the audio path whatsoever. As for impact on the main voice, the other (unique) feature is the ability to split the main voice and the sibilance to separate channels. In this mode of operation there is absolutely nothing being done to the audio path at all - total transparency! The only thing being done is to switch the output (smoothly) when sibilance is detected. When there is sibilance, the audio path is fed to the sibilance output, when there is no sibilance it is fed to the main output. There is no threshold, so it doesn't matter if the sibilance is soft or loud - it always works. If you split the signal with the DeEsser and leave both channels alone, the signal will null with the original when inverted because nothing is being done to the audio other than switching between outputs!
All to say that at that point, you can do whatever you want to either channel without affecting the other. Brighten up the main vocal without affecting sibilance. Turn the sibilance down a bit or EQ it so it's not as harsh (saturation can also be useful on sibilance to smooth it a bit in some cases). Add reverb to just the main voice, keeping the sibilance out of the reverb to avoid "splatty" reverb sibilance (or add a tiny bit of sibilance to the reverb so it's not totally unnatural). You can automate the sibilance level if you want TOTAL control over the amount of sibilance, useful for those tricky voices.
It sounds like this level of control is what you're looking for - have you had the chance to trial the Selig DeEsser?

[guitfnky]

anyhoo, the problem is, if I get to a phrase where I'm really pushing the vocal, that starts to hit the de-esser as well, so it's doing the usual thing any gate does when it's not triggering the way you want--causing problems.

I know that's sort of a thing with most every other de-esser, so I don't expect to eliminate the problem completely, but I'm looking for ideas/best practices on how I can minimize the impact of the non-eSSeS. how are pro de-esser plugins designed to handle that? does it take a lot of additional processing (what I'm getting at is, is it possible to do using stock devices).

any advice would be super welcome!

That's the problem with any dynamics based de-esser - sibilance will be reduced based on how loud it is rather than whether or not it's present. You would think this would be want you want, and much of the time it works. But if you have a section of a song where the vocal gets louder/quieter overall, the de-essing won't work well because it's reduction is based on it's overall level rather than relative to the main voice. Meaning, if you set the threshold to de-ess the loudest sibilance, it will reduce the sibilance relative to the main voice by a certain amount. But then if the voice gets soft, the sibilance will ALSO be soft and thus will fall below the threshold. So in that part of the song the sibilance won't be reduced at all, and the voice/sibilance relationship will remain unaffected.

Most of the time a dynamics based de-esser is built by simply feeding the EQ'ed (isolated) sibilance frequency to the side chain of a compressor/limiter. Alternatively, a crossover/splitter can be used to process just the sibilance range, ducking it when energy exceeds the threshold in the sibilance range. Another issue with this approach is that sibilance doesn't always occur at the same frequency, even over the course of one song with one voice (let alone different signers on different songs). So you have to either feed a wide band of frequencies to the detector or automate a narrow band to follow any changes. This approach can also miss other constant sounds that may need reducing such as F, T, and Ch sounds.

What exactly do you mean by "eliminate the impact of the non-esses? What do you hear happening to the main voice (non-ess) currently?

Shameless self promotion mode: ON!
Of course I HAVE to mention the Selig DeEsser here, which was developed precisely to work with situations like this. As for transparency, it doesn't do ANYTHING to the audio path except for a level adjustment - no filters are in the audio path whatsoever. As for impact on the main voice, the other (unique) feature is the ability to split the main voice and the sibilance to separate channels. In this mode of operation there is absolutely nothing being done to the audio path at all - total transparency! The only thing being done is to switch the output (smoothly) when sibilance is detected. When there is sibilance, the audio path is fed to the sibilance output, when there is no sibilance it is fed to the main output. There is no threshold, so it doesn't matter if the sibilance is soft or loud - it always works. If you split the signal with the DeEsser and leave both channels alone, the signal will null with the original when inverted because nothing is being done to the audio other than switching between outputs!
All to say that at that point, you can do whatever you want to either channel without affecting the other. Brighten up the main vocal without affecting sibilance. Turn the sibilance down a bit or EQ it so it's not as harsh (saturation can also be useful on sibilance to smooth it a bit in some cases). Add reverb to just the main voice, keeping the sibilance out of the reverb to avoid "splatty" reverb sibilance (or add a tiny bit of sibilance to the reverb so it's not totally unnatural). You can automate the sibilance level if you want TOTAL control over the amount of sibilance, useful for those tricky voices.
It sounds like this level of control is what you're looking for - have you had the chance to trial the Selig DeEsser?

yep, you've pretty much nailed my conundrum with your penultimate sentence. the sibilance in my voice has definitely been tricky. switching mics has helped a lot, but I still have recordings of using those other mics that are really good takes I don't want to rerecord. I have tried a number of de-essers (I didn't just demo it, I own yours already ), and they all get me like 75-80% there, but that last 20% or so sounds very unnatural, or doesn't do enough, and that's what's been driving me crazy. I HATE having to automate something that's supposed to be automatic--stupid point of principle, I acknowledge.

I actually demoed Soothe 2 for this purpose (which people seem to swear by), and it did a pretty good job on a few of the tracks, but I found it was still lacking, in certain cases, it was making the esses sound just...weird--difficult to describe. it did de-harsh the sound, but it also didn't sound natural. almost as if the esses were still there, but still had a lispy quality to them. needless to say, I didn't drop the large chunk of cash needed to purchase that.

it's hard to know what it is about most of the normal de-essing plugins that doesn't seem natural to me, but I suspect it's a combination of:
- the attack and release timing--sometimes it sounds great, but in the cases it's not working so well, it REALLY doesn't work well, and/or
- the need for it to perform a wideband de-essing based on an obviously very narrow essing band -- for my voice at least, it seems that de-essing only the essing frequencies is either wildly inconsistent, or consistently sounds bad

the inverted phase thing I've got going is already working on the wideband signal by design, so that's the easy part. and so far, somehow it doesn't seem to have that first problem, either. I'm guessing that's something to do with how the gate timing settings are 'releasing' the inverted phase signal, and how the resulting volume increase in that signal interacts with the un-inversed version--just seems more natural. I do still need to try it on a few more tracks to see if it's really doing what I want.

and I have noticed the issue you describe, with using dynamics-based de-essing. what I've got (and here's that caveat again) so far seems to trigger well as long as I set my threshold properly, and I'm finding that for the few esses it misses, it really isn't necessary anyway.

when I said 'eliminate the impact of the non-esses', I just meant basically what you described--when the highly filtered vocal used to identify the esses gets loud enough (I'm singing louder), those pitched sections are becoming loud enough in the ess-ranges of the detector that they end up tripping the gate threshold. that was introducing the inverted signal into the middle of sung sections, causing unwanted volume dips. the compression trick Timmy mentioned seems to be working well, to move stuff out of the way of the ess 'sensor', although I did have to add a control to allow that to be adjusted, if there are still 'false positives' coming through, causing it to trigger inappropriately.

[guitfnky]

and not a small part of the inspiration for this is simply curiosity—I’ve never heard of an inverted phase de-esser, so wanted to see if it was something that’s possible/practical.

[selig]

yep, you've pretty much nailed my conundrum with your penultimate sentence. the sibilance in my voice has definitely been tricky. switching mics has helped a lot, but I still have recordings of using those other mics that are really good takes I don't want to rerecord. I have tried a number of de-essers (I didn't just demo it, I own yours already ), and they all get me like 75-80% there, but that last 20% or so sounds very unnatural, or doesn't do enough, and that's what's been driving me crazy. I HATE having to automate something that's supposed to be automatic--stupid point of principle, I acknowledge.

I can only say that 100% of the time de-essing sounds unnatural is because you're doing too much. That's exactly where the "lisp" effect comes from - when you hear that, you've gone too far!

I can also say that I've never had to automate the Selig DeEsser except for in one very extreme case I was trying to get the sibilance totally removed from every line 100% clean, which is no small feat. And fwiw, it only took automating a very few places to achieve that. I was testing the device and pushing it hard to see what it could do, and it passed with flying colors.

Again, the Selig DeEsser doesn't affect anything except level, and you can look at the results (by printing to an audio track) and see that even in extreme cases it is fast enough to not affect the main voice at all, only targeting the sibilance. Here's a quick test I suggest. First, split the sibilance to it's own channel starting with a dry vocal, and solo each to be sure you're in the right mode etc. Now lower the sibilance channel fader, and notice how it sounds - keeping in mind you are doing absolutely nothing to either the vocal or the sibilance except for changing the level of the sibilance. This is actually exactly what happens inside the DeEsser - if you choose "6 dB Reduction" it's exactly the same as splitting the sibilance and lowering the sibilance fader by 6 dB! Which is to say, anything "odd" about the sound of such an arrangement is 100% down to the oddness inherent in the ideal of sibilance being lower than it naturally would be, and 0% down to any "processing" side affects or artifacts. Make sense?

If you want additional feedback, I'd be happy to take a look at one of your "trouble maker" vocals and show you how I'd deal with it.

[selig]

and not a small part of the inspiration for this is simply curiosity—I’ve never heard of an inverted phase de-esser, so wanted to see if it was something that’s possible/practical.

The end result is not that different than compressing the sibilance, the only variables being the time constants and amount of gating/gain reduction. BTW, if you are interested in using polarity inversion for effects, this is similar to doing the "send ducking" trick, where you use a send and an inverted audio signal to apply ducking to any number of devices without having to connect them via cables etc. You can also build a transient shaper using similar techniques - lots of fun with simple concepts!

[guitfnky]

yep, you've pretty much nailed my conundrum with your penultimate sentence. the sibilance in my voice has definitely been tricky. switching mics has helped a lot, but I still have recordings of using those other mics that are really good takes I don't want to rerecord. I have tried a number of de-essers (I didn't just demo it, I own yours already ), and they all get me like 75-80% there, but that last 20% or so sounds very unnatural, or doesn't do enough, and that's what's been driving me crazy. I HATE having to automate something that's supposed to be automatic--stupid point of principle, I acknowledge.

I can only say that 100% of the time de-essing sounds unnatural is because you're doing too much. That's exactly where the "lisp" effect comes from - when you hear that, you've gone too far!

I can also say that I've never had to automate the Selig DeEsser except for in one very extreme case I was trying to get the sibilance totally removed from every line 100% clean, which is no small feat. And fwiw, it only took automating a very few places to achieve that. I was testing the device and pushing it hard to see what it could do, and it passed with flying colors.

Again, the Selig DeEsser doesn't affect anything except level, and you can look at the results (by printing to an audio track) and see that even in extreme cases it is fast enough to not affect the main voice at all, only targeting the sibilance. Here's a quick test I suggest. First, split the sibilance to it's own channel starting with a dry vocal, and solo each to be sure you're in the right mode etc. Now lower the sibilance channel fader, and notice how it sounds - keeping in mind you are doing absolutely nothing to either the vocal or the sibilance except for changing the level of the sibilance. This is actually exactly what happens inside the DeEsser - if you choose "6 dB Reduction" it's exactly the same as splitting the sibilance and lowering the sibilance fader by 6 dB! Which is to say, anything "odd" about the sound of such an arrangement is 100% down to the oddness inherent in the ideal of sibilance being lower than it naturally would be, and 0% down to any "processing" side affects or artifacts. Make sense?

If you want additional feedback, I'd be happy to take a look at one of your "trouble maker" vocals and show you how I'd deal with it.

part of my problem is that, at least for me, sibilance is very much a 'you can't unhear it' kind of thing. I tend not to notice it...until I do, and then it's something that drives me kind of batty, in ways that no other aspect of mixing ever really do.

I've just done a lot of playing around with both my de-esser (which I've taken to calling the InverSSor ), and yours, and I've definitely noticed some interesting stuff--I can't say I have any real conclusions yet, but the tests were very interesting. here's some of what I did/noticed:

- listening to just the Sibilance on your device, and doing the same on my combi, I noticed for yours, the detection, and switch back are almost instantaneous--mine has all the speed of a fairly fast SSL gate, and the release is definitely noticeable when you're listening to it. yet, when you apply aggressive de-essing (purposefully taking it into hard lisp territory), it doesn't seem to be cutting out the audio as quickly as if you're just listening to the Sibilance setting--is there an internal crossfade happening in Selig DeEsser? (this is more out of curiosity than anything)
- in one test, I went by ear, setting both de-essers just to the point where they started to sound unnatural, then I backed off until each one sounded natural again--in that test, which was over audio with just a single (loud) instance of sibilance, the Selig DeEsser was able to get the ess down to -23.6 db (all measurements below done @ LUFS, using Youlean Loudness Meter 2), and the InverSSor was able to get it down to -24.3 db
- in a variation of the same test, I did the same thing, but for an entire phrase, with multiple different levels of sibilance--in that one, DeEsser was down to -19.4 db, and InverSSor down to -20.2 db

at that point, I thought my InverSSor was doing pretty well (and I guess I still think so, considering it's my first attempt at making a de-esser), but I kept going...because I want to use a de-esser as a 'set-and-forget' device as much as possible, I did the same basic test, but over a larger section of music--set it so I could hear something just starting to go 'wrong', and backing off each de-esser.
- after setting the overall maximum tolerable de-essing for this entire section, I tested one of the louder esses--here, the DeEsser took the ess down to -17.2 db, and the InverSSor only took it down to -16.6 db
- I then left the two at the same settings, and retested using a quieter ess--the DeEsser got down to -21.2 db, and the InverSSor, down to -21.0 db

obviously, that's in no way scientific, and I have no idea how useful LUFS may be for measuring sibilance, but as I said, I thought the results were interesting. maybe worth noting; I left the DeEsser in D-S mode, with the crossover at the default 2khz, the V-S Sensitivity at the default of 0, and no lookahead for all of these tests.

I'm still playing around with them both, but so far still leaning slightly more toward the InverSSor for my own voice (although they're closer than I expected). there's just a slightly different character to the two, I think.

I'll probably have the combi dialed in enough to post it tomorrow, and I'll try to do a quick video at some point. after that first version is done, I may play with another way to accomplish the same thing, but instead of using a gate, I might try to convert the audio's volume to CV to ride the line mixer's volume (for the inverted signal)--I'll still have to figure out a way to gate it before it gets translated into a CV value though somehow, I think (or gate the CV value?), because there will still be level passing whenever there's singing, and it obviously wouldn't make sense to just have low-level cancellation happening all the time; then it's a volume reducer, not a de-esser!

re: your other response--I definitely love the idea of messing around with inverted signals. you can do super-interesting stuff with them, and as you say, the concept is really simple!

[guitfnky]

okay, so here's the InverSSer combi.

connect the ins and outs of the combi as you see fit (since it's a combinator, it won't fit inside an existing rack chain, but it can still be routed in series)

it has these controls:
- gain reduction indicator on the Channel Dynamics -- use this to see when the de-esser is being triggered; if it's not triggering as you want, you need to twist some of the knobs and press some of the buttons below
- focuSS -- raise this to focus the ess detection more towards the upper frequencies (if the device is triggering when it shouldn't be, on sung notes, instead of esses and other sibilant sounds)
- leSSen -- raise this to cut more of the sibilance out, and get that lovely little lithpy thound; lower it for a less aggressive de-ess sound
- threSShold -- controls the level at which the sibilance will trigger the de-esser; set it to where it's de-essing the stuff that's bugging you, and not the stuff that isn't
- DEACTIVATOR -- this mutes the inverted signal; effectively same as using the Bypass toggle
- liSSten -- mutes the NON-phase-inverted signal, so you only hear what the sibilance detection is doing (as long as the DEACTIVATOR button isn't also depressed; if it is, the entire signal will be muted)
- Gertie -- this does nothing but look pretty
- Sally -- this too does nothing but look pretty
- Ted -- Ted is useless also; he's not very attractive, but is an okay guy

I'll post a video shortly, and maybe put it in its own thread.

of course, definitely love any feedback, suggestions, or "hey, that's stupid, and here's why" constructive criticism.

[selig]

A great test of "set it and forget it" is sending the same vocal line at different levels and seeing how each respond. This will most likely reveal which can truly be set and forget…one feature of my de-esser is the ability to detect sibilance at any level, when it's softer than the main vocal or when it's louder and everything in between. You can therefore run it either before or after other processing, such as compression. In face, using my Leveler I can 100% level a voice and the de-esser works exactly the same. Dynamics based de-essers struggle when the sibilance approaches the same level as the main voice, for obvious reasons - sibilance then has a similar threshold level as the primary vocal, making it difficult to set the threshold.

An extreme case is when de-essing a very bright vocal, where the frequency range of the sibilance is as loud or louder on the main voice as the sibilance. In the most extreme cases a threshold based de-esser cannot accurately differentiate between sibilance and breathy vocals!

Another test would be to see if even amounts of de-essing are applied in all cases, and whether or not your system can detect other consonants as well as breaths. With your system, all sibilance that passes the gate is inverted and causes some cancellation with the original. The louder the sibilance, the more cancellation. While you would think this would work best, I find the sibilance to inversely track the main voice, meaning that the louder the main voice gets, the softer the sibilance gets. Thus, soft lines have relatively louder sibilance, and loud lines have relatively softer sibilance, which sounds unnatural to my ears. With my system, sibilance is reduced by a consistent amount - but the beauty is that you can compress the sibilance in extreme cases (because it can be on it's own channel) and thus control sibilance more naturally if it's not consistent to start, yet another thing you cannot do with other systems.

You are also lucky - you are designing your device for one voice, yours. I used a collection of 20-30 different vocals (from projects over the years) to ensure it worked exactly the same on each and every voice/type. So it's likely you can eventually end up building a de-esser that works almost as well as mine for your voice because you won't have to account for any edge cases. And while building things that already exist is a great way to learn, I would also encourage you to begin exploring building things that don't already exist!

EDIT: in the time I was writing this you posted your Combinator, which I'll take a look at in another post when I get a chance!

[guitfnky]

A great test of "set it and forget it" is sending the same vocal line at different levels and seeing how each respond. This will most likely reveal which can truly be set and forget…one feature of my de-esser is the ability to detect sibilance at any level, when it's softer than the main vocal or when it's louder and everything in between. You can therefore run it either before or after other processing, such as compression. In face, using my Leveler I can 100% level a voice and the de-esser works exactly the same. Dynamics based de-essers struggle when the sibilance approaches the same level as the main voice, for obvious reasons - sibilance then has a similar threshold level as the primary vocal, making it difficult to set the threshold.

An extreme case is when de-essing a very bright vocal, where the frequency range of the sibilance is as loud or louder on the main voice as the sibilance. In the most extreme cases a threshold based de-esser cannot accurately differentiate between sibilance and breathy vocals!

Another test would be to see if even amounts of de-essing are applied in all cases, and whether or not your system can detect other consonants as well as breaths. With your system, all sibilance that passes the gate is inverted and causes some cancellation with the original. The louder the sibilance, the more cancellation. While you would think this would work best, I find the sibilance to inversely track the main voice, meaning that the louder the main voice gets, the softer the sibilance gets. Thus, soft lines have relatively louder sibilance, and loud lines have relatively softer sibilance, which sounds unnatural to my ears. With my system, sibilance is reduced by a consistent amount - but the beauty is that you can compress the sibilance in extreme cases (because it can be on it's own channel) and thus control sibilance more naturally if it's not consistent to start, yet another thing you cannot do with other systems.

You are also lucky - you are designing your device for one voice, yours. I used a collection of 20-30 different vocals (from projects over the years) to ensure it worked exactly the same on each and every voice/type. So it's likely you can eventually end up building a de-esser that works almost as well as mine for your voice because you won't have to account for any edge cases. And while building things that already exist is a great way to learn, I would also encourage you to begin exploring building things that don't already exist!

EDIT: in the time I was writing this you posted your Combinator, which I'll take a look at in another post when I get a chance!

awesome, thanks, definitely appreciate any feedback you might want to give. oddly enough, I think the ability to use a threshold might be (part of) why I like what I've got here. sometimes the sibilance is low enough that it doesn't need to be (or indeed shouldn't) be reduced further. at least that's what I've found with my own vocals. I hadn't thought of using the separate out and using a compressor on the sibilant channel--that's something I'll definitely have to try!

my setup definitely captures other consonants as well. I've found that (again for my own voice) this sounds just fine, and sometimes improves some of these non-esses when I hadn't even noticed they might be poking out.

and you're totally right, I'm being very selfish in my approach (and I don't mean that as a slight against myself )--it seems to work very well for me, so far, but to your point, I'm only really testing with my own voice, using a few different microphones. I would LOVE for it to be useful to someone else (hence sharing it), but I definitely don't have lines from 30 other vocalists to pull in to test, so it may be completely useless for most others.

I do have to admit though, I find it kind of funny (and take a certain amount of pride) knowing that--at least to my ears--this combinator that I spent a few days hacking together with Reason's stock devices is outpacing the $200 tool I was trialing for that specific purpose. it really goes to another point you've made many times elsewhere--there is no quick-fix tool. that plugin (Oeksound Soothe) is really good at de-harshing, but it gets a lot of praise for use as a de-esser as well, and in that regard, it's certainly useful, but also just as hit-and-miss as most other de-essers, in my experience. and of course my InverSSer is far from perfect--it's just been working well for me, and any way you cut it, that's pretty cool.