Howto: make peak reducing automation?

[111]

So here`s an issue. One profi mastering guy during a short conversation about final loudness of mastered track advised me several things but here`s the one I don`t understand a little bit - to make an automation of peak reducing instead using regular limiter and that should give about 3 db`s or so to raise on every track in which this technique is used without any bad impact on it...soooo, how can I do this?)I`m awared with automation as itself but still don`t know how to make it working.

[Fergler]

That's bulltshit, think of what you have to do for this:

a) find the highest peak somehow.. assuming there's only one of them, there's probably hundreds of varying height
b) draw in automation of volume
c) raise the gain of the entire track to replace the headroom you just gave yourself from reducing that peak.. assuming it is the only peak. You still have to limit it now to avoid clipping because all those tiny peaks you ignored are now the new highest peaks because of the gain increase. Which would be just fine if you were using automatic limiting but not if it's manual. It's a vicious circle

A limiter does this for you. Simply set the threshold and make sure there is look-ahead (LoudMax has this) so it's transparent. The attack/release should be 0ms to act as a 1:1 fader control.

Then push the gain into the limiter until you only see gain reduction on the peaks.

[111]

Well, ok, I suspected something wrong in it) So you suggest using look-ahead limiters, right? I`m familiar with loudmax, it hasn`t these knobs such as attack/release, etc. Could you advise another one with them? Or any look-ahead will suit? And how can I see if only peaks are reduced? Cause I see a stripe showing that somethings`s reducing, and that`s all about it. Thanx!

[fixerdave]

hi, go look here, might be what you want dave
http://forum.cockos.com/showthread.php?t=143741

[planetnine]

Or, of course, you can wonder why the hell you are wasting time researching for the express purpose of mangling a (presumeably) well-mixed piece of music to raise its intrinsic loudness by a couple of dB in an age where i-tunes, Foobar, broadcast sources, etc all will turn it back down again because consumers don't like the levels jumping all over the place..?

Or you could leave it with those transients intact to give the track a bit of punch to stand out from the peak bludgeoned sheep-products on offer?

Look up loudness normalisation. People will look back on the last decade or so and wonder what they were thinking (and where the original mixes are archived).

Just an idea.



>

[Quest The Wordsmith]

111, Reaper has a stock JS plugin that does this already: JS: LOSER/ZeroCrossingMaximizer. What it does is it takes a peak, finds the zero crossings of that peak (before & after), then lowers that peak to the threshold. I used to do this manually in GoldWave (an audio editor) for small files, like kicks and snares. Reducing just the single peak would let me raise the whole volume substantially. Loser's JS plug does this on the fly. There is notable latency introduced, so this isn't for live mixing or anything. I have mine set to 500ms.

[111]

Thanx!JS: LOSER/ZeroCrossingMaximizer seems to be a good thing, I`ve turned knobs a little bit, very similar to plain look-ahead limiters but maybe different. GainRider may be something interesting too but it supposed to be for a vocal and hardly I can give "decent feedback" as I`m unsophisticated user=) Thanx anyway. I really can`t understand why should the one maximise track killing dynamics instead of just using volume knob of monitors/headphones/amps/player by the listener... but they demand.

[Fergler]

just fyi loudmax doesn't have attack/release because attack/release isn't used (is set to 0) in lookahead master limiters of that design. The name is LoudMax, maximum loud. That is how you get maximum loud. If you had attack of anything other than 0ms the peaks would 'peak' through and clip. If you have release then the volume damping will affect audio that is not peaking which will cause pumping.

The look-ahead is to avoid distortion that is common in real-time 0ms compression. You can see that in action by loading reacomp and putting it's attack at 0.

Zero crossing JS plugin is the same idea

[111]

Ok, now I understood, will work further. thanx!

[Quest The Wordsmith]

For the record, Loser's ZeroCrossingMaximizer isn't the same as limiting. Limiting, technically, is extreme compression. Compression changes the waveform very subtly with the ratio's. Loser's JS plugin just volume-reduces the peaks with no audible discernment since it does it at the before and after zero crossings.

[111]

Yeah, I think that`s what I really need on that stage!

[ashcat_lt]

It might not take much to get that Zero-Crossing Maximizer to spit out some form of control signal. Maybe it could wiggle a slider which could be recorded as automation somehow, or if nothing else it could spit out MIDI CCs, though that wouldn't be quite sample accurate. It could, however, be edited afterward, if you really wanted to get tweaky about it.

[111]

Maybe)editing afterward however isn`t too convenient as I told about single tracks, not master. As far as i understood Zero-Crossing Maximizer should do what I need without additional automation. Thanx anyway, I`ll keep in mind that possibility!

[Fergler]

Originally Posted by Quest The Wordsmith

For the record, Loser's ZeroCrossingMaximizer isn't the same as limiting. Limiting, technically, is extreme compression. Compression changes the waveform very subtly with the ratio's. Loser's JS plugin just volume-reduces the peaks with no audible discernment since it does it at the before and after zero crossings.

Any change in volume is a change in waveform. So what are you referring to here that is different?

Plugins like LoudMax do exactly as you describe. The moment a sample goes over the threshold level, let's call it -2 for e.g. anything that goes above -2db is clipped off at -2 and because it's attack/release of 0 it reacts to every sample (hence the need for look-ahead).

[ashcat_lt]

Maybe I'm missing something, but I don't really see how lookahead helps without something similar to an attack/release mechanism. If you're just clipping off samples one at a time anyway, then why do you need to know that you're going to do it ahead of time? The only reason to look ahead is to keep from changing the gain too fast in order to avoid adding too much distortion at the output. It says "I'm at gain of 1 now, and I have to get down to 0.5 and I've got so many samples to get there. Better start heading that way now!" Otherwise why introduce any more latency than necessary?

ReaComp with it's lookahead, rms size, and knee parameters is actually a pretty flexible tool for this kind of thing. I am intrigued by this Zero Crossing Maximizer, though. I'll have to bust it out next time I need to knock down some pokeys.

[111]

Gentlemen,right now I have another question, it`s pretty closed to the first issue, so I won`t make many threads=). You see, I`m looking through a course of mixing metal (in fact style doesn`t matter here). The "teacher" does it in Protools. One thing I really liked - clip gain. Can`t understand how he really did it... whith that feature he turned down by 3 db every track, then normalized to 0 db and again turned down by 3 db using that feature, and all tracks became louder (in comparison with tracks without those manipulations) without visible changing of waveform and audible distortion.... ̶W̶h̶a̶t̶ ̶t̶h̶e̶ ̶h̶e̶l̶l̶.̶. How can smth similar being achieved in my lovely Reaper? Sorry for my dullness, but I really have a feelling that I`m missing smth important...

[ashcat_lt]

That's just frickin' weird!

I'm sure there's more going on there. Got a link?

As stated it's like:

Theoretically, this is a form of distortion/compression, but kind of from the bottom up. You will get some measurable RMS increase by increasing the denormal and quantization noise relative to the signal, but in a floating point environment it would be barely noticeable, and even in fixed point I would imagine it would just sound like more noise, if anything.

[Quest The Wordsmith]

Originally Posted by 111

Gentlemen,right now I have another question, it`s pretty closed to the first issue, so I won`t make many threads=). You see, I`m looking through a course of mixing metal (in fact style doesn`t matter here). The "teacher" does it in Protools. One thing I really liked - clip gain. Can`t understand how he really did it... whith that feature he turned down by 3 db every track, then normalized to 0 db and again turned down by 3 db using that feature, and all tracks became louder (in comparison with tracks without those manipulations) without visible changing of waveform and audible distortion.... ̶W̶h̶a̶t̶ ̶t̶h̶e̶ ̶h̶e̶l̶l̶.̶. How can smth similar being achieved in my lovely Reaper? Sorry for my dullness, but I really have a feelling that I`m missing smth important...

Without seeing the video, this sounds a lot like what I would achieve with ZeroCrossingMaximizer. I'm pretty sure the guy is processing the sound as opposed to on-the-fly processing. But What it seems is happening is the peaks are being reduced by 3 db, thereby creating headroom. Then, the tracks are normalized to zero and he repeats again. In the end, he has 6 db of gain increase.

To achieve the same thing with the JS plugin, just put the threshold down to -6 and your done. It's doing the exact same thing.

[serr]

Originally Posted by 111

So here`s an issue. One profi mastering guy during a short conversation about final loudness of mastered track advised me several things but here`s the one I don`t understand a little bit - to make an automation of peak reducing instead using regular limiter and that should give about 3 db`s or so to raise on every track in which this technique is used without any bad impact on it...soooo, how can I do this?)I`m awared with automation as itself but still don`t know how to make it working.

What he's saying is that in his opinion/experience, MANUALLY riding the volume on a track is preferable (cleaner, better sounding) than using a compressor or limiter to reduce the peaks in question. And of course you can automate volume fader moves with automation envelopes.

Try it yourself and find out!

The advice in this post of OTHER compressors or limiters to try is all good stuff but misses the point of your mastering guy's claim (that manual volume riding is better than ANY compressor/limiter). Some people are purists about this and truly prefer to spend hours drawing or recording automation for riding the volume fader instead of strapping a compressor on the channel.

I don't like the 'pillow in front of the speaker' effect from a cheap compressor either! But when you need to control dynamics in a mix... you need to do something. Leaving something wild in a mix would make any snobby talk about compressors a moot point. And if you have the time and hunger to manually draw volume automation... well, great!

[111]

Damn, I really didn`t understand him at 1-st time. It turned out that all his tracks for some reason were too silent since beginning and using a clip gain function he manually turned them up. He also used something like analyzer of gain plugin to show the current level so that he could fit all the tracks to -3db level for a headroom. Overall I asume that clip gain is something similar to a plain volume automation but what`s the reason to use analyzer of gain - that`s still mystery for me. Maybe only because it can analyse track which is not played... The course is called Groove 3 Mixing Heavy Rock Metal, it`s paid so a link is to nowhere)

I wish I`ve come across whith a full and good course about heavy rock\metal in Reaper)

[serr]

Clip gain would be equivalent to item gain in Reaper.

Why would you want to adjust the gain there instead of the channel volume or channel pre-fader/fx volume?
Simply workflow.
For film sound for example you might have bits of audio from wildly different sources. It's really quick to spot gross level differences visually and it's simply quick to use this method to adjust them.

When you're mixing music and you just have a few fiddly adjustments to make here and there in the song that are not isolated to a particular item - then track volume automation makes more sense to use.

Why would someone want to have a quick glance at max peak level from a gain analyzer instead of running said audio clip a few times in real time while watching the peak meter? Because of that!

[111]

Ok, I`m persuaded now that`s a good feature (gain analyzer). Any vst analogs?

[serr]

There's an SWS action: "SWS: Analyze and display item peak and RMS"

Using a (VST, AU, or other) plugin and watching the meter would be the "manual" way I described.

Install the SWS actions and extensions. There's a link on the Reaper page.
You can assign a keyboard shortcut if you intend to use this often. (I assigned shift-P for example)

[Fergler]

Originally Posted by ashcat_lt

Maybe I'm missing something, but I don't really see how lookahead helps without something similar to an attack/release mechanism. If you're just clipping off samples one at a time anyway, then why do you need to know that you're going to do it ahead of time? The only reason to look ahead is to keep from changing the gain too fast in order to avoid adding too much distortion at the output. It says "I'm at gain of 1 now, and I have to get down to 0.5 and I've got so many samples to get there. Better start heading that way now!" Otherwise why introduce any more latency than necessary?

ReaComp with it's lookahead, rms size, and knee parameters is actually a pretty flexible tool for this kind of thing. I am intrigued by this Zero Crossing Maximizer, though. I'll have to bust it out next time I need to knock down some pokeys.

It's just an aspect of digital processing. Dynamically processing so many samples is taxing on the system. I believe the distortion part comes as a result of 'inter-sample peaks' but I could be wrong.

Anyway, try it yourself and see. Open up reacomp with default settings, set attack and release to 0, ratio to inf, and keep threshold at 0db. Now push the audio into it to achieve limiting - it distorts wildly.

Lookahead just gives enough time for those distortion artifacts to be circumvented by having enough time to properly process each sample individually ahead of time before it has to play them back.

No idea, btw, whether this remains true during rendering in non-realtime. Never tried it.

[ashcat_lt]

Originally Posted by Fergler

It's just an aspect of digital processing. Dynamically processing so many samples is taxing on the system.

No. Straight clipping is just a ceiling function. "If the input is bigger than x, make it x". It's super efficient. "Soft" clipping applies a curve, and some of those can be a bit intense to calculate, but in no case is the distortion caused by "taxing" the system. If the CPU can't keep up, you'll get dropouts or glitches, not what you're describing.

I believe the distortion part comes as a result of 'inter-sample peaks' but I could be wrong.

No. Inter-sample peaks only cause distortion on playback if they ask the DAC to produce a voltage that it cannot. Try the thing with ReaComp that you describe below, but drop the Output slider to like -12 so that there's plenty of headroom for those inter-sample peaks. You still hear the distortion.

Anyway, try it yourself and see. Open up reacomp with default settings, set attack and release to 0, ratio to inf, and keep threshold at 0db. Now push the audio into it to achieve limiting - it distorts wildly.

I have. In fact, with all of it's different parameters (especially the knee control and up to 64x oversampling), ReaComp is a pretty cool little distortion engine. I haven't actually recorded anything with it in this capacity, but I did play around for a while with an instance of ReaEQ on either side of ReaComp and came up with a pretty convincing and flexible stompbox emulation.

Lookahead just gives enough time for those distortion artifacts to be circumvented by having enough time to properly process each sample individually ahead of time before it has to play them back.

No idea, btw, whether this remains true during rendering in non-realtime. Never tried it.

The distortion is caused by the sudden change in gain, period. It happens in analog as well, and yes it's the same in both real-time and offline modes.

The only way that lookahead can help is if you actually use it to predict the amount of gain reduction you need at the peak, and then gradually head toward that on the samples around the actual one that you want to squash, so that the resultant waveform is closer to the original with less "sharp edges" than might be caused if you just clip off the one sample. It must either work on an RMS averaged envelope or apply some sort of attack/release routine. The actual A/R times might be hidden from the user, they might be fixed or somehow program dependent, or "built into" the RMS averaging, but they must be there if you want to reduce distortion.

[111]

Originally Posted by serr

There's an SWS action: "SWS: Analyze and display item peak and RMS"

Using a (VST, AU, or other) plugin and watching the meter would be the "manual" way I described.

Install the SWS actions and extensions. There's a link on the Reaper page.
You can assign a keyboard shortcut if you intend to use this often. (I assigned shift-P for example)

Thanx! I`ve found and already tried. It shows everything pretty good. Unfortumately when I try to edit gain to amplify it to -3db level (through media item properties menu) unlike that training video tracks begin sounding horrible and "clippy", even if metering shows everything is OK...although not all of them Unstable technologies...

[Fergler]

Originally Posted by ashcat_lt

No. Straight clipping is just a ceiling function. "If the input is bigger than x, make it x". It's super efficient. "Soft" clipping applies a curve, and some of those can be a bit intense to calculate, but in no case is the distortion caused by "taxing" the system. If the CPU can't keep up, you'll get dropouts or glitches, not what you're describing.


No. Inter-sample peaks only cause distortion on playback if they ask the DAC to produce a voltage that it cannot. Try the thing with ReaComp that you describe below, but drop the Output slider to like -12 so that there's plenty of headroom for those inter-sample peaks. You still hear the distortion.


I have. In fact, with all of it's different parameters (especially the knee control and up to 64x oversampling), ReaComp is a pretty cool little distortion engine. I haven't actually recorded anything with it in this capacity, but I did play around for a while with an instance of ReaEQ on either side of ReaComp and came up with a pretty convincing and flexible stompbox emulation.


The distortion is caused by the sudden change in gain, period. It happens in analog as well, and yes it's the same in both real-time and offline modes.

The only way that lookahead can help is if you actually use it to predict the amount of gain reduction you need at the peak, and then gradually head toward that on the samples around the actual one that you want to squash, so that the resultant waveform is closer to the original with less "sharp edges" than might be caused if you just clip off the one sample. It must either work on an RMS averaged envelope or apply some sort of attack/release routine. The actual A/R times might be hidden from the user, they might be fixed or somehow program dependent, or "built into" the RMS averaging, but they must be there if you want to reduce distortion.

Enlightening, thanks. Explains why the LOSER/transientcontroller and sleepy-time DSP Transient both go haywire at pushed settings (unuseable distortion).

[111]

So, how can I quickly normalize tracks to -3db or another exact level I want instead of trying to get it whith the line of gain aujustnment in a media item properties menu (not convenient)? Maybe somewhere it can be printed: Normalize to __db? Volume faders isn`t the same...