Everyone knows from the analog world that compressors are not created equal. Setting the knob to ten and getting two has been a common joke in audio for decades. and this joke is almost entirely about compressors. Is this the case with inexpensive and free VST compressors people are using on their streaming and podcast projects?
Before looking at tone and fineries, basic function should be evaluated, just to know that the knobs and control settings are meaningful at all, and if a compressor can even handle the task at hand.
Might as well find out.
I grabbed a series of inexpensive, free and at least one “normally priced” VST compressor and made up a terribly flawed test.
Everything would be set to 4:1 ratio if possible. A nice even number and often as high as would be used on dramatic vocals. Hard knee if possible too.
5ms attack to see if we can beat the 15ms good for vocals, and a 50ms release. Fast enough to see any weirdness and problems
Then, an -18dB threshold to correspond with common voice actor and book reader recording targets. I know people go below this and use over easy features to bring up low parts, but this is not necessary just to know if the compressor can do anything useful.
I ran a -18dBFS tone through all the VST compressors with settings as close to this as possible. Recorded the output level.
added channel gain if necessary to reach -18dB due to undefeatable soft knee or other nonsense.
Then increased the tone level to -6dB, a typical peak target for voice actors, and recorded the output level presented. It is easy from here to interpolate the actual compression ratio compared to the stated 4:1 setting.
Observing a completely vertical transition between -18dB and -6dB gives us at least some idea of the nature of the attack and release.
Finally, a shaped noise output with the same energy profile of voice recordings was also tried to see if there was any difference in behavior between tone and noise, but the results were much the same with all the compressors tested.
Most of all, trying not to take this too seriously.
01 LALA (limiter setting)
-18dB input tone -18dB threshold (should have no effect, return -18dB)
Soft knee. Needs +3dB to return to -18
-6dB input tone, ratio 4:1 (should be -15dB)
-18 (limiting correctly)
Attack speed (no control available)
13ms
Release speed (no control available)
1s extremely long curved release slope.
02 JA 1175 FAST
-18dB input tone -18dB threshold (should have no effect, return -18dB)
-18dB ok
-6dB input tone, ratio 4:1 (should be -15dB)
-15dB ok
Attack speed Error in test. Set to 5μs
Instant. Zero leak through. So it really is fast.
Release speed set to 50ms
Steep curved release slope. steep until 8ms and flattens out by 220ms
03 FirComp
-18dB input tone -18dB threshold (should have no effect, return -18dB)
-18dB ok
-6dB input tone, ratio 4:1 (should be -15dB)
-15dB ok
Attack speed set to 5ms
7ms
Release speed set to 50ms
Steeply curved release slope. Sharp return until 0.12s and reaching 0db at 1s
04 LALA compressor
-18dB input tone -18dB threshold (should have no effect, return -18dB)
soft knee add 3dB at fader
-6dB input tone, ratio 4:1 (should be -15dB)
-15dB ok
Attack speed (no control available)
10ms
Release speed (no control available)
1s release extremely long curved release slope. (no controls and same as limiter setting)
05 M Compressor
-18dB input tone -18dB threshold (should have no effect, return -18dB)
-18dB ok
-6dB input tone, ratio 4:1 (should be -15dB)
-13dB
Attack speed set to 5ms
3ms quite fast
Release speed set to 50ms (0.05s)
Linear release 0.1s
06 RoughRider 3
-18dB input tone -18dB threshold (should have no effect, return -18dB)
-18dB ok
-6dB input tone, ratio 4:1 (should be -15dB)
-13dB
Attack speed set to 5ms
4ms, ok.
Release speed set to 52ms
Linear release 61ms
07 JS Express Bus
-18dB input tone -18dB threshold (should have no effect, return -18dB)
-18dB ok
-6dB input tone, ratio 4:1 (should be -15dB)
-13dB
Attack speed (no control available)
10ms
Release speed (no control available)
Curved release slope. 0.16s sharper and tapers to 0.5s
08 ReaComp
-18dB ok
-18dB input tone -18dB threshold (should have no effect, return -18dB)
-6dB input tone, ratio 4:1 (should be -15dB)
-12dB (more like 2:1)
Attack speed set to 5ms
5ms ok
Release speed set to 50ms
Linear 100ms
Notes: I have been told the ReaComp release slope is “anti-log”
09 FabFilter Pro-C 2
-18dB input tone -18dB threshold (should have no effect, return -18dB)
-18dB ok
-6dB input tone, ratio 4:1 (should be -15dB)
-12dB (more like 2:1)
Attack speed set to 5ms
4ms ok
Release speed set to 50ms
highly curved release slope. 20ms fast, then gradual past 1/3rd second
10 T compressor at 50
-18dB input tone -18dB threshold (should have no effect, return -18dB)
-18dB ok
-6dB input tone, ratio 4:1 (should be -15dB)
-11dB (more like 1.5:1)
Attack speed “fast”
13ms
Release speed “fast”
10ms curved release slope. Very fast.
11 T compressor at 100
-18dB input tone -18dB threshold (should have no effect, return -18dB)
-18dB ok
-6dB input tone, ratio 4:1 (should be -15dB)
-9dB (more like 1.25:1)
Attack speed “fast”
3.5ms fast
Release speed “fast”
3ms curved release slope. Super fast.
12 JS Master Tom
-18dB input tone -18dB threshold (should have no effect, return -18dB)
soft knee or mismatch add 6dB
-6dB input tone, ratio 4:1 (should be -15dB)
-7dB (Almost no effect. I do not know how to drive this contraption. Likely entirely “on the knee.”)
Attack speed (no control available)
Difficult to evaluate. 0.17s
Release speed (no control available)
Difficult to see. Release steps up slightly 0.35 seconds after release event.
I also ran some signals through the most inexpensive hardware compressors I could find out how usable this would be. Something that would fit the same target audience as the free and affordable VST crowd. Interesting discoveries.
DBX 1046
Ratio set to 4:1
Threshold set to 0dB (this is analog)
Attack speed set to (No control available)
attack 36ms
Release speed set to (No control available)
release linear 0.36S
Limiter set to +5dB
soft clipping first 3ms
DBX 166XL
Ratio set to inf:1
Threshold set to 0dB
Attack speed set to full “fast”
Attack 7ms
Release speed set to full “fast”
release curved 0.275s
Limiter set to 0dB
well controlled clipping first 3ms More square and hard than 1046 limiter.
Behringer 2600 compressor
Ratio set to 4:1
Threshold set to 0dB
Attack speed set to 0.3ms
Attack 13ms first 3ms gets through
Release speed set to 0.1s (100ms)
Release linear 25ms
Behringer 2600 compressor and +6 limiter
Ratio set to 4:1
Threshold set to 0dB
Attack speed set to 0.3ms
Attack 18ms
Release speed set to 0.1s (100ms)
Release linear 25ms
Limiter set to +6dB
first 4ms hard clipping
Many of these cannot manage a 5ms attack time, however it is important to know that this arbitrary 5ms attack speed is not necessary with typical human speech. It’s super overkill.
If you just look at some vocal recordings from random reads, you will notice that vocal sounds do not appear instantly. No matter how fast they seem, they appear gradually over several wavelengths.
Here are a couple of random specimens measured in seconds. The length of the rising slope at the beginning of words.
0.017
0.043
0.036
0.021
0.019
0.013
0.075
0.018
0.043
0.016
0.020
0.029
0.029 seconds is of course, 29ms (milliseconds, or 29/1000ths of a second.)
You can see here the fastest in this random pile is 13ms, with many much slower, with an average here of 29ms.
So it is safe to guess that 100ms is too slow and will let peaks through at the beginning of each sound. Something useful with some musical instruments but not really with spoken word. Also, 15ms attack is more than enough speed for vocals and for the most part, anything below 10ms is unnecessary.
you can see from these tests that the number on the dial is often not accurate, yet at some setting , it is possible for most or all of these to do the job.
Of course, none of this address the TONE of the compressors or their reactions to different frequency distributions, distortion etc. but merely their ability to compress a generic mixture and mid band tone as indicated.
I also did not revisit these tests to see if more appropriate settings could be found. Most interesting example being the T-Compressor with the fast release speed being very speedy indeed.
But this is not a controlled experiment as much as an exploration to see if more tests are interesting and worth conducting.
Turns out, they are. More later.