Continuing from the first two episodes, we will now examine the behavior of the limiter in all three threshold configurations: center, below, and above. We will also perform some audio normalization and observe which of the limiters applies the strongest compression for those who prefer heavy audio limiting. Dynamic audio limiting is a process used in audio production to control the volume of a signal by setting a maximum threshold. When the audio exceeds this threshold, the limiter reduces the gain to prevent distortion or clipping. Unlike compression, which reduces dynamic range more gradually, limiting focuses on preventing peaks from going above a set level, ensuring consistent and controlled output.

We will skip the explanation of how to read these graphs and what each column and row in this graphics above represents, since we have explained that part in the first episode. Let’s start with the middle row. This is how a typical limiter is supposed to work: after a set threshold, a signal is limited to the desired ratio. In this case, we used 1:100. Once again, we can observe strange behavior on the downward slope. Why this happens, we may never know.

In either case, I think we can conclude that for limiting functions, an external limiter should be used, as we are clearly losing around 3 dB of volume. Right after the audio passes the midpoint in time, the limiter starts applying some extra limiting, dropping the audio from -10 dB down to -13 dB. Afterward, it slowly recovers; however, during this time, strange ripples are produced, as if two functions are fighting each other. Eventually, the audio recovers to -12 dB, losing around 1 dB of volume, and then, after the threshold is passed, the limiter disengages. Here is a full-resolution picture to better illustrate what’s happening:

Ideally, none of what happens in the second part of the audio should occur. This waveform should be completely flat, as it is in the first part of the recording. The reason for this is, as we explained in the first episode, a source was a constant sine wave at a constant pitch was used. There shouldn’t be any of these rippling or volume loss effects.

Moving back to our table above. Looking at the first row, we see the most extreme type of limiter the Emulator 4 provides. Of course, this is created using the threshold center setting. It will literally flatten everything. Use this if you really want to sound loud, although, as many would say, loud = boring after a while.

Looking at the bottom row, as expected, since the threshold is in the ‘below’ configuration type, the affected audio is below our threshold point. Use this to dirty up drum samples recorded in a live environment, as it will keep the original transients while exaggerating all the background noise. Let’s now try the same but with extreme settings.

As we can see (just for fun) I decided to push the limiter to the extreme (a threshold at 70% is something you will rarely use) just to see if the problem with volume loss still occurs—and it does. In the middle row, we can see an inaccuracy in the restoration of the original signal. Although the area is below the threshold, it is still, for some reason, undergoing compression. The two sloped lines (representing the descending stage of the audio) should be parallel, but they aren’t.

And last but not least, let’s see, once we normalize the audio, which compressor configuration is the ‘strongest,’ with the least amount of unfortunate volume loss. Looking at the results, there is no clear winner. Although the third row looks the best with the least volume loss, we need to keep in mind that it’s compressing a huge amount of audio (70%), which is not something we would do in everyday use.

This concludes our exploration of the compressor effect in the Emulator 4 series. It was something I always wanted to do, as I’ve always found the results sounded a bit odd at times. Some of the findings confirm that, but on the other hand, they also inspire us to try some of the more exotic options available and hear the results in more common applications, such as guitar, drum, or vocal sounds. These have much more dynamic content compared to static white noise or a constant-pitch sine wave, which we used here to determine the overall envelope of the compressor/expander curves. It was an interesting journey, in any case.

In the last episode we have learned that E-MU Emulator 4 gives us three distinct compressors, rather than one. Sharp-eyed readers likely noticed that we didn’t mention one feature of the compressor—a setting that provides two additional options and is related to the compressor analysis mode, which can be either Peak or RMS based. This wasn’t an omission but rather a deliberate choice to avoid making the text too lengthy—and potentially tedious to read. However, in the case of a compressor, limiter and expander, using the correct mode can be a critical setting worth understanding.

RMS (Root Mean Square) compressors measure the average power or energy of the audio signal over time, which closely aligns with how humans perceive loudness. Instead of responding to short, sharp peaks, they consider the overall “energy” of the signal. Characteristics:

• Smoother, more natural compression.
• Ideal for leveling overall volume rather than reacting to transient spikes.
• Often used for vocals, bass, and mix bus applications where consistent loudness is desired.

An example usage would be reducing the overall loudness variation in a vocal track without reacting to quick peaks like plosives.

Peak compressors respond instantly to the highest signal levels or transients, regardless of the overall energy of the signal. They focus on capturing and controlling sharp, sudden peaks. Characteristics:

• Precise and reactive.
• Ideal for controlling sharp transients and preventing clipping.
• Often used for instruments with pronounced attacks, such as drums or percussive elements.

An example usage would be taming the attack of a snare drum or preventing distortion in a highly dynamic recording.

Armed with all this knowledge, let’s now get back to our Emulator 4 and run some more tests. This time, we will test the expander in all three threshold configurations: Below, Center, and Above.

The Expander
A dynamic expander is an audio processing tool used to increase the dynamic range of a signal by amplifying the differences between loud and soft parts. Instead of reducing the range by lowering louder signals, an expander decreases the volume of sounds below a set threshold, making quieter sections even softer relative to the louder parts.

This behavior is useful for noise reduction, as it can lower background noise levels when the main signal (such as speech or music) falls below the threshold. Expanders can operate in downward expansion, where the quiet parts get quieter, or upward expansion, where louder parts are made even louder. This is where our Emulator 4 comes into play as it offers all two options, plus additional one called “Center” which is in a way the combination of the two.

The key parameters of an expander include the threshold, which determines when the expansion begins, and the ratio, which controls the degree of expansion – for example 1:2 ratio doubles the dynamic difference. Unfortunately Emulator 4 does not provide us with rational numbers for the Expansion control (e.g.,1:2, 1:3, etc.) instead is uses Real numbers like 0.50:1 which in this particular case would be equal to 1:2 expansion ratio.

We will skip the explanation of what each column and row in this graphics above represents, since we have explained that part in the first episode. Now let’s start with the third (bottom) row, as it uses the expander in a threshold configuration set to “Below.” This is how your average expander is supposed to work. At around -6 dB, it starts applying dynamic expansion, and everything seems to look correct. However, when we examine the falling slope (the second half of the waveform), we can notice a strange feature: the expansion is not symmetrical.

In fact, it brings us back to the compressor “issue” or feature that we mentioned in the first episode. While I can understand a compressor having such a feature, I can’t figure out why this would be done for an expander, which is primarily a tool for “repair” rather than adding character. What’s happening is that the expansion on the falling slope is not being applied as it should. It misses almost 3 decibels of audio before starting the expansion in that part of the audio. One can’t help but notice the similarity in behavior to the compressor. But then, we have to keep in mind that this is the same compressor algorithm, which I guess explains this “feature”. I know the image is very small for detailed inspection, but in the high-resolution version, I can clearly see how the expander misses the -6 dB mark and starts expanding at around -9 dB. A bug? I guess we’ll never know.

Now, let’s examine the middle row. Remember when we mentioned earlier that some expanders make loud portions of the audio louder? That’s exactly what the expander in the “Above” threshold setting does. As we can see, it doesn’t affect any audio until -6 dB, after which it starts expanding, making the middle of the waveform (the peak) even louder. In this case, it results in overdrive. Therefore, I’d recommend never using normalized (maximized) audio in this configuration, as it will definitely produce distortion/overdrive.

Finally, in the top row, we have a rather exotic type of expansion—a sort of “extreme” expander. This mode combines the previous two methods by attenuating audio below -6 dB and applying gain above -6 dB. This could have potential for percussive or drum-type sounds buried in heavy noise, as this expander effectively removes that noise. Again, be very careful with the gain control, and don’t normalize the audio before using it in this configuration. Here is another example of expansion with different settings:

This time, the original audio was set to -1 dB to prevent clipping phenomena for the two expander threshold types that make loud parts louder. Slightly stronger expansion was used this time, with a 1:2 ratio at -6 dB. In hindsight, I should have set the source to -3 dBFS for even better preservation of the peaks in the middle of the wave (the loudest portion), but when zoomed out so far, it would have been difficult to see other details. So, I opted for this approach.

Let’s start with the third row. Again, the expansion is not symmetrical—something isn’t right in the descending part of the audio. I also noticed “leakage” in the low-energy area, as if the expander didn’t entirely reduce the volume in a 1:2 ratio but suddenly shifted to something closer to a 1:1.1 ratio (or less). I’m not sure why this happens; perhaps I’m missing something.

The vertical scale in all these graphs is logarithmic, which means the displayed volume reduction should appear linear—but it doesn’t. The other two threshold options behave as expected: the “Above” setting makes the area above the threshold louder, while the “Center” threshold option once again shows us a rather exotic, extreme expansion taking place. I guess one could create quite sharp percussive sounds using this threshold configuration. In the next episode, we will look into a limiter and conduct some tests with normalized audio.

Among its many tools, the E-MU Emulator 4 features a compressor, expander and limiter for processing audio samples. The manual explains their usage, although not in the most straightforward way. It also includes some rather exotic settings for the threshold in three modes: Above, Center, and Below. The best way to understand how they work is to see them in practice. The explanation in the manual describes their behavior as follows:

• Above: Only signal levels above the threshold % will be affected by the compressor.
• Center: Signal levels above and below the threshold % will be affected by the compressor.
• Below: Only signal levels below the threshold % will be affected by the compressor.
• The % determines the threshold level as a percentage of 100% of 16-bits.

My understanding is that only in the position labeled Above does the compressor behave in the way we’re accustomed to. In the Below configuration, it acts as an inverse(?) compressor, which can be a bit hard to grasp, while in the Center configuration, it seems to function as a combination of the two. I know it’s confusing, hence let’s move to the graphics instead.

The explanation from the User manual might satisfy some, but it still leaves a few questions unanswered, especially when we factor the Expander into the equation. Instead of trying to explain this in detail, the best approach is to display the results graphically. Each row in the table below shows the settings that were used and the resulting output. Although the source signal is the same, it’s displayed alongside the processed result in every row for clarity, which is why it repeats. In the first series of tests, white noise was used at 0 dB FS, and in the second series, a sine wave was used at -1 dB FS. In all tests, a 10 second long waveform was used.

The Compressor
Let’s explain what’s shown in the image below. There are three columns. On the left, we have the Emulator 4 screen display with the compressor (or expander) settings. In the middle (shown in green), we see the source waveform, which starts at -96 dB, fades in all the way to 0 dB, and then fades out back to -96 dB during the period of 10 seconds. It’s literally a heavy zoomed out waveform display of a simple fade-in and fade-out waveform designed to reveal the envelope of the compressor (and expander). In the right column (shown in green), we have the processed waveform, again zoomed out heavily. Layer below it, in gray, is the outline of the input waveform, making it easier to see how the compressor’s envelope affects the signal. Since we have three threshold types (Center, Above, Below) we have three rows.

Threshold type: Above
Let’s first look at the middle row, which represents the threshold type setting: Above. This setting is how your average (normal) dynamic compressor works. Let’s now focus on the settings that were used. With the threshold set to 50%, it seems to activate at -6 dB, which I guess is correct. Still, it’s a pity they didn’t use a dB scale, as working with percentages is very tricky. It essentially involves trying to manipulate a logarithmic scale using a linear tool. Unfortunately, it is what it is, and we can’t change that.

The ratio was set to 5:1, and we can see the gain reduction falling somewhere around -5 dB. This seems reasonable, as for every 5 dB input, we should get a 1 dB increase. Using this calculation, it’s approximately 6:5, or around 1.2 above 6 dB. Apologies for the image being small (there were simply too many tests / images involved to place at the same page). Looking at the vertical axis on the right side of the graph: the top tick is 0 dB, the second tick is -3 dB, then -6 dB, -12 dB, and so on. So far, everything seems correct.

However, examining the envelope shape, it seems that something is “off”. Either there’s an error in design (which I doubt, as E-MU was kingpin back then), or the compressor was modeled after a very specific type of design. Unfortunately, the manual doesn’t provide enough detail to clarify. We observe a nonsymmetric response during the descending (fade-out) stage, which shouldn’t happen. The signal should be compressed symmetrically in both directions, but instead, on the descending slope, it rapidly loses volume, then pauses in amplitude briefly before almost morphing into expansion.

This behavior is indeed strange, and I’m trying to understand what this design might be modeled after. It’s certainly not a classic opto compressor—it must be something else entirely. Keep in mind that this test uses a 10-second long sweep. Just imagine such a drastic change of volume on short sounds like kicks and snares. We also need to note how the compression seems to continue even after the waveform has reached its peak and passed below the threshold. In fact, compression occurs throughout most of the sample, from the middle to the end, at what appears to be a 5:1 ratio. This is easily noticeable, even on this small image—just observe the angle of the descending slope of the processed signal and compare it to the original, which is shown grayed out in the background. All in all, it’s a rather intriguing and unconventional design!

Threshold type: Center
Let’s now look at the top row, which represents the threshold type setting: Center. This appears to be a compressor that works with the average signal value, seemingly combining the behavior of the second and third rows, which represent the Above and Below threshold settings, respectively. In this configuration, it seems to automatically gain the volume below the threshold while compressing the volume above it. This results in a rather interesting type of compression.

Threshold type: Below
Next, let’s examine the bottom row, which represents the “Below” threshold setting. As expected, it behaves as described—it leaves the area above the threshold untouched while compressing and auto-gaining the area below the threshold. This is quite an exotic and unique compressor design.

Such functionality could be particularly useful, as it maintains the integrity of transients while adding gain to the quieter portions below the threshold. Honestly I’ve never encountered a compressor like this before—it must be incredible for drum and percussive sounds!

Gentle compression
In the image below, we have another example of compression—this time with a mild 2:1 ratio set at 50%. The numbers appear to be correct, although the output should not exceed -3 dB. Between -6 dB and 0 dB, there should only be a gain increase of 3 dB, resulting in an endpoint exactly at -3 dB, rather than the observed -2.8 dB (hi res picture shows peak at around -2.8dB). However, let’s not forget, this slight discrepancy could be a feature of the particular (analog modelling) design being emulated here.

Conclusion
It seems that E-MU has provided us with three distinct compressors hidden under the hood, each offering completely different results and envelope shapes. Some are excellent for “rounding” the waveforms, while others are entirely exotic types of compression. In the next episode, we will explore the similarly intriguing behavior of the expander and limiter, providing tests with normalized data, among other things, and determine which compression technique produces the loudest sound—a trend that seems to be popular these days (though not among the author).