General Concepts for Controlling Dynamic Range, aka Crest Factor

Have an urge to learn, or a calling to teach? Want to share some useful Youtube videos? Do it here!
User avatar
selig
RE Developer
Posts: 9449
Joined: 15 Jan 2015
Location: The NorthWoods, CT, USA

Post 21 Jul 2015

This is a response to a question Josh Mobley asked me privately, and it was suggested I share it somewhere where others could see it because it may address issues that arise from time to time. So naturally I figured why not share it here?!? ;)

Originally Posted by neoverse
I have a small question. I posted this thing on Facebook about your -12 thing but now, I have another question if you don't mind.

If I level a track down to -12 but it's the transient that has the most volume ( a big ride in this case) and the true average volume in rms is like -25 db (after the -12 db reduction) how do i manage that? compressing? limiting? Expansion?

all these years in the biz and mixing is still some black magic voodoo shit.

thanks in advance!

**************************************************************************************

To answer your question, the concept I will discuss here is known either as "short term Dynamic Range", or the term I prefer which is "Crest Factor". https://en.wikipedia.org/wiki/Crest_factor

Crest Factor
Basically, the crest factor is the difference between the peak and the average level of a waveform. This is very easy to see in Reason by using the VU + Peak meter mode. For some sounds, such as raw drum recordings, the crest factor can be as high as 20-30 dB. Most typical sounds have a crest factor between 10-20 dB. For reference, a sine wave has a crest factor of 3 dB, a triangle wave is 4.77 dB, while a square wave is the ONLY waveform that has a crest factor of 0 dB, the technical limit for a waveform.

The square wave is the exception because it spends no time between the values of 100% and -100%, creating an average voltage of 100% (since waveforms are rectified when measuring level, meaning negative values are counted as positive). Thus, the average level is exactly the same as the peak level, and the difference (and therefore the crest factor) is "zero". Every other waveform, whether theoretical or real world, will have a non-zero crest factor since there will always be some difference between their peak and VU/average level.

Loudness
When looking at more complex waveforms, such as an entire mix, crest factor can be useful to judge overall perceived loudness. In Reason this is easily done by using the "VU + Peak" meter mode. Personally speaking, I tend to shoot for a crest factor on most pop/rock/dance tracks roughly around 12 dB. With complex waveforms the crest factor will vary over time, so you will be looking at the "typical" (or even the maximum) crest factor to judge loudness when using this technique. When using a brick wall limiter such as the Maximizer or Ozone, the peak levels will typically remain close to 0 dBFS, which allows you to use only the VU level to read an accurate crest factor value. Since the peak level is zero, the crest factor is simply the VU level (more accurately, the absolute value of the VU level, since crest factor is always a positive value). This simply means that with the peaks hitting 0 dBFS consistently because of limiting, the VU level (ignoring the minus sign) IS the crest factor - simples, right? For example, if the VU is -12 dBFS (assuming the peak is limited to 0 dBFS), the crest factor is 12 dB. If the VU is -15 dBFS, the crest factor is 15 dB, etc.

Techniques for Increasing Perceived Loudness
Now on to your specific example. With a peak level of -12 dBFS, and a VU/RMS of -25, you have a crest factor of 13 dB, which is VERY manageable. Again, a kick or snare drum, when initially recorded, often has a crest factor over 20 dB or more, and this is one reason we typically seek to reduce it for a modern/loud mix. Note that many drum samples already have some dynamic processing, thus often lowering the Crest Factor and making further processing potentially less necessary. By combining the techniques mentioned below, you can increase the loudness while reducing the chance of making the process "obvious" or audible to the listener. The available approaches include traditional downwards compression (assuming a fast enough attack time), as well as upwards compression (useful for not squashing transients). There is also any number of saturation/distortion techniques which themselves are a form of dynamic range reduction (read: limiting). I find saturation effects to be the most transparent on signals with high transient levels/crest factors, especially of you use parallel techniques for your saturation effects (though it's not always necessary to do so to achieve the benefits).

Overview
Besides the processes themselves, there are a few ways to combine these processes to further achieve your goal of loudness without artifacts. I'll cover each of the processes below, as well as the common techniques used to combine them.

Compression/Limiting
This is the most common approach to reducing the crest factor of an audio signal. The only guideline here is to remember to use a fast enough attack time or you'll actually INCREASE the crest factor (which can be useful, but is not what you were asking about!). If the attack time cannot be set fast enough, as in the case of the SSL channel compressor, you will actually increase the dynamic range/crest factor, which in many cases may be desirable because it can add attack or "punch" to the track. This is especially helpful on individual tracks such as kick and snare, but you may want to follow it with one or more of the techniques I outline below to control the overall peak level of your mix. BTW, this is a very popular sound I still find a use for today even though it's over 30 years old! Note: on the SSL channel compressor, even when using the "Fast" button the attack is only 3ms per 20 dB, which is still not fast enough to effectively reduce dynamic range/crest factor. OTOH, a compressor like the FET (1176 clone) and others have a fast enough attack time to effectively control crest factor, but be careful not to "give away" the process by over doing it. With compression, a little can go a long way in my experience. I suggest starting small and adding more gain reduction only if needed. A few dB of compression, especially when high ratios are used, can be enough to get the job done in many cases. Remember, there's no "free lunch", and sometimes it's best to use multiple stages of subtle compression to achieve your goals as opposed to trying to get one compressor to handle the whole job. See "Serial Compression" below for more on this technique.

Saturation
One of my favorite approaches to increasing loudness, either parallel or "direct", is any number of saturation type effects, especially those that seek to emulate the effects of the big three "T's: tape, transformers, or tube devices. This is mainly due to the fact they offer a "one-two punch". What I mean is they accomplish two things at once: they increase the average level by effectively clipping the peaks to some degree, and they add harmonics which increases the "density" of the track and therefore the overall perceived loudness. As a bonus you also gain the ability to better perceive low frequency sounds like bass and kick on smaller speakers, mainly due to the higher harmonics being added or increased in level. This effect can work best with drums since we are not very sensitive to short term distortion or clipping, as long as it occurs for less than 5-10 ms of the duration of the signal. Luckily, the peaks/transients we seek to reduce in drums and percussion instruments are often very short in duration, allowing a greater amount of saturation/clipping to occur before it's audible. When combined with parallel techniques, the effect can be especially subtle, with the added benefit of being able to independently EQ the saturated channel to reduce the energy in the frequency range where it is most audible (typically in the 2-6 kHz range, as we are most sensitive to these frequencies: see "Fletcher-Munson Curves" for more information: https://en.wikipedia.org/wiki/FletcherMunson_curves


Here are some of the typical ways to apply the above processes to control Crest Factor/dynamic range of individual instruments, sub-mixes, or the entire mix/master.

Direct Processing
Though this term isn't commonly accepted, I'm using it here to contrast the other techniques I will cover below. This is the most common type of processing, also known as "insert" processing. This is simply the technique of inserting a dynamics or saturation based device directly into the signal path, and thus affecting the entire signal. This approach can work quite well in many cases to reduce dynamic range/crest factor, but only if you have a devices with a fast enough response time as mentioned previously. It's worth noting that when combined with other processing such as EQ, or other saturation and/or dynamics effects, the order of processing can make a big difference - be sure to try changing the order if you are not liking the results you are getting!

Serial Compression
This is a specific form of "direct/insert processing, involving multiple compression devices in series. One general rule I've found is that you don't need to totally crush the track in question with a single device when using multiple processing stages, as each stage only need to compress by a small amount to achieve the overall effect. Serial compression has long been done, more often in ways that are not obvious at first. For example, if you compress a recording when tracking, and then compress again when mixing, then again when mastering, the resultant track you hear has gone through at least three compression stages serially! If it's a drum track, such as a kick or snare, you may have compressed it 4-6 times by the mastering stage: once when tracking, agin when mixing (on the individual channel, on the drum bus, on the master bus), and again during mastering (which may even involve two stages of compression when using "glue" compression and "brick wall" limiting). One general guideline to using serial compression is that it is often more beneficial to first lightly compress and then more aggressively limit (in that order). Mastering is the obvious example of when you want to use a brick wall limiter as the LAST processor in the signal chain in order to more strictly control the final peak level and avoid clipping. For other applications, YMMV, which means you may get a better result by using two compressors or two limiters in a row, or you may find that limiting first achieves the desired result. As always, don't be afraid to experiment in order to choose the best approach for the task at hand. Note: a "limiter" is generally assumed to simply be a compressor with fast attack and high ratio.

Parallel Compression
This popular technique can also help to reduce dynamic range/crest factor, and therefore increase loudness, but it takes a slightly different approach. For parallel compression to be effected at reducing Crest Factor, and therefor increasing loudness, there are a few guidelines to follow. The main guideline is that it is essential to use a fast enough attack time on the parallel track processing in order to create a very dynamically "dense" signal. This "concentrated" signal is then blended in with the original in small amounts to create the desired final crest factor. You can think of the original track as the "peak" or "transient" channel, and the compressed channel as the "average" channel. This approach (IMO) works best on sub-mixes rather than individual tracks, and helps control any extra attack/transients you may add with the SSL channel comp or similar. While it turns out it is in many cases very similar to simply using a lower ratio "insert" compressor, there are other variables that can be introduced on the parallel channel (like EQ or saturation) that will definitely differentiate it in practice. Personally speaking, I often find it easier to control the amount of compression using a parallel approach rather than an insert approach.

Dry/Wet vs Parallel Channels
Adjusting the balance between the original and the effects signals allows you to easily control the resultant crest factor. This is commonly done in one of two ways; with a blend or "dry/wet" control, or with a separate level control for the original (dry) and effects (wet) signals. I prefer the "two channel" approach to parallel processing over the "dry/wet" approach in most cases, since I don't want the original track to be reduced at all - I only want to add the "average" signal. Meaning, a dry/wet control will reduce the dry level as it increases the wet, and most compressors with this control don't have a way to start with both wet and dry signal at the same peak level (shameless plug: my Leveler RE is an exception to this rule, as it is specifically designed to overcome these limitations in other devices). This means that with other devices, a setting of 50% does NOT always mean there are equal amounts of dry and wet signal, which can make it more difficult to tweak compressor settings as you will also have to tweak the dry/wet control to maintain the same balance.

Mixing it Up
With parallel or serial processing you can of course substitute saturation for compression, or even combine dynamics and saturation based effects (as well as EQ/filtering) to great success. As always, the results you get can vary based on the processing order, so don't forget to explore the options if you're not getting the expected or desired results with the first approach.

Summing it Up
Ultimately, giving real world (and audio) examples are often far more helpful for questions such as this one, and I'm more than happy to do so here. But I hope this in depth post covers the subject in enough detail so that you (or anyone else reading this) can understand the basic concepts and begin to apply these techniques in your mixes to great "effect".
:)
Selig Audio, LLC

lowpryo
Posts: 452
Joined: 22 Jan 2015

Post 21 Jul 2015

loved this post! I like using saturation for the character that it adds, but I never thought about how it actually does increase perceived loudness because it makes the signal more harmonically dense

could you provide insight on your favorite way to apply parallel saturation in Reason? I've tried using Scream's Tape mode in parallel and I get some weird phasing issues

User avatar
Olivier
Moderator
Posts: 1248
Joined: 15 Jan 2015
Location: Amsterdam

Post 21 Jul 2015

Great read. Thanks Selig :)
:reason: V9 | i7 5930 | Motu 828 MK3 | Win 10

User avatar
selig
RE Developer
Posts: 9449
Joined: 15 Jan 2015
Location: The NorthWoods, CT, USA

Post 21 Jul 2015

lowpryo wrote:loved this post! I like using saturation for the character that it adds, but I never thought about how it actually does increase perceived loudness because it makes the signal more harmonically dense

could you provide insight on your favorite way to apply parallel saturation in Reason? I've tried using Scream's Tape mode in parallel and I get some weird phasing issues
Scream's Tape mode has an issue that is (mostly) addressed by inverting the polarity when using it in parallel. I say "mostly" because it's actually not exactly 180° out of phase. :(

For saturation, I mostly use Saturation Knob on parallel channels along with Normen's VMG-01 (with a 4 sample delay) on the dry channel. I will sometimes use Pulveriser, but there are plenty of other RE options these days if these options don't float your boat! ;)
Selig Audio, LLC

lowpryo
Posts: 452
Joined: 22 Jan 2015

Post 21 Jul 2015

nice! one more question while it's on topic. do you use any particular device for measuring VU level, or do you just scroll up to the big meter? I use your Gain to get the peak on pretty much every track, but I hate repeatedly scrolling up to the big meter when I'm setting up my compressors. (and I know, I should really be going by ear first and foremost. but I'm learning that meters are really helpful!)

User avatar
Benedict
Posts: 2379
Joined: 16 Jan 2015
Location: Gold Coast, Australia

Post 25 Jul 2015

Thanks Selig. Great article and info

:)

User avatar
selig
RE Developer
Posts: 9449
Joined: 15 Jan 2015
Location: The NorthWoods, CT, USA

Post 26 Jul 2015

lowpryo wrote:nice! one more question while it's on topic. do you use any particular device for measuring VU level, or do you just scroll up to the big meter? I use your Gain to get the peak on pretty much every track, but I hate repeatedly scrolling up to the big meter when I'm setting up my compressors. (and I know, I should really be going by ear first and foremost. but I'm learning that meters are really helpful!)
Speaking strictly for myself…
I really only care about the actual crest factor "value" when mixing/mastering, and the big meter is excellent for this IMO and all I've ever needed.

When I approach dynamics processing in a mix, I'm doing one of two things generally speaking: increasing dynamic range (adding attack/punch), or decreasing it (adding loudness/sustain). The end results should speak for themselves - either the chosen processing creates the desired end result or it doesn't. So in that light I don't need to know about the actual crest factor of each individual track or sub- mix, I only need to know whether the processing addressed the issue or not, and if not I'll adjust things until it does.

IMO that's the whole point of adding any processing during a mix - it should be done to address an issue you are HEARING. As such, it should be obvious by listening whether the issue has been successfully addressed or not. Visual tools are excellent at locating or providing further details about the specific issue you are HEARING. But I have hardly ever found them all that useful to identify whether there IS an issue or not in the first place (clipping is one exception to this rule, as it's not always easy to identify occasional clipping when you are focusing on the more creative aspects of a mix)!

Examples where visual tools/meters help me:
Something sounds over-compressed, so I check the gain reduction meters.
I hear a resonant frequency dominating a recording, so I check a spectrum display (this one is also easy to address by sweeping EQ).
I hear digital clicks in a recording, so I zoom in to visually find and eliminate them.

Examples where visual tools/meters aren't necessary for my work:
The track or mix has too much low/mic/high frequency energy.
The mix is muddy (or similar).
The kick or snare isn't hitting hard enough.
An individual track isn't "present" enough in the mix.

Hope that clarifies the issue, as it was not my original intention to suggest using crest factor values as guides for all dynamics related processing. I only wanted to introduce the concept of crest factor and demonstrate ways of how to control it one way or another in a mix situation. :)
Selig Audio, LLC

Ostermilk
Posts: 1520
Joined: 15 Jan 2015

Post 26 Jul 2015

Great read Selig.

It might look like a long read for some but I reckon you've covered the ground completely and very succinctly.

Kudos, excellent post.

  • Information
  • Who is online

    Users browsing this forum: CommonCrawl [Bot] and 0 guests