Creative Audio Processing
07 - Processing tools : dynamics, compression

 

 

 

Compression is a very complex phenomenon, quite hard to understand and really master. Let aside the basics, compression depends heavily on the machine that's used - much more than it is with EQs. Compression can be a major tool when it comes to give style to a production.

 

1. Compression basics   Originally, compression is strictly a dynamic tool : it reduces the dynamic range of the source. For example, if I've got a source which RMS power varies from -40dB FS to 0dB FS, I can decide that the goal RMS power will be varying from -40dB FS to -20dB FS, and then I boost the level by 20dB, now the RMS power will be varying from -20dB FS to 0dB FS. The processed sample will have a more constant level, for the same peak level. It can mean two things depending on the observation's scale. On a large time scale, there will be less pianos and less fortes. Soft parts of the music will be closer to loud parts. On a close time scale : sounds will likely to be "thicker", with a subjective aspect closely depending on the compressor that's used. This last point is very relevant to our subject - it's a style related issue.

 

2. Attack time - release time   Attack time and release time are two very important compression parameters. We've seen that a compressor is able to detect high levels from a source and to get them closer to low levels. Let's suppose our compressor receives low levels for some time. Suddenly, there is a very loud transient coming. The compressor is supposed to suddenly add a negative gain to the sound. Now, if the compressor adds an immediate gain (attack time = 0 samples), what will happen ? Such a sudden gain can result in a "drop" in the waveform. And a drop in a waveform, that will sound like a digital click, and it's not good. The compressor must apply the gain more smoothly, meaning the gain correction should come more slowly. That's where the attack time is important : it determines how smoothly the gain will be applied. Now the release time : same principle. How smoothly does the compressor get out from the "gain" zone ? When the compressor stops receiving high levels, how much time passes until the compressor stops applying the gain ?

 

3. Attack and release times : consequences on perception and grain   Let's consider the following situations : sudden high level transient, different attack times, and see what the consequences on perception of different attack times. NB : once more, actual perceptive result much depends on the actual machine, but basic principles remain true whatever the compressor. - Very short attack time ( <= 1ms) Level is very quickly "readjusted", meaning that there will be no or few electrical overload - on the other hand, the attack may sound "crushed". - Short attack time (below 20 ms) "Readjustment" will take a while. 20ms is still under the ear's integration time, meaning that the listener will not be able to clearly realize what's going on, but the percussive aspect of the sound will remain. On the other hand, some unwanted or "unmastered" distortion can happen : several milliseconds will still be above target level. That's where the compressor brand or type or technology is also important : distortion is likely to happen inside the machine (after the compression stage), and resulting aspect will depend of how the compression can handle inner overload. - Long attack time (>50ms) Readjustment is above the ear's integration time, so the listener will have a clearer idea of what's going on. Unless this is deliberate, it's not necessarily a good idea : most of the time, we want to modify the timbre and grain, not literally telling the listener that we have a compressor on the audio chain. In most cases, if distortion doesn't happen, this kind of settings is at the very least peculiar, and more generally useless, at least when it comes to "transient handling" ; the transient is likely to last less than 50ms. If distortion does happen, then it's a different : advantage can be taken from the compressor's natural distortion.   Now let's consider different release times. Release times are generally higher than attack times. - Short release time (<50ms) Release time is under the ear's integration time. Low levels following the transient will sound immediately higher. This kind of settings can bring a very thick sound : everything is always loud. - Long release times (1 second....) When dealing with this kind of settings, the machine's intrinsic technology is important. What kind of "upwards curve" will the gain adjustment follow ? How will it get back to 0 ? Compressor are psychoacoustically adjusted, so that the listener will not be able to hear the transition from a lower level to a higher level. However, with some compressors, it doesn't really work, and listeners are likely to experience the famous "pumping" effect. More generally, the "upward level curve" is an important element in the compressor's "subjective/grain" aspect. - Very long release times (>2/3 seconds) When release times get very long, the "upward curve" shape is not so important. The ear is less able to follow such a long evolution, at least not in a complete mix. The curve's shape will be less critical, and the pumping effect less likely to be heard. There may be an exception to this last remark : extremely precise compressors, such as the Massenburg stereo unit, are quite difficult to handle even with a long release time. But then, precise is not necessarily musical.

 

4. Summary : grain / design / subjective aspects and machine type   We've seen that compressor settings are important when it comes to modify the sound's aspect. We've also seen that the machine itself is important, whatever the settings are. Let's sum up that last point. - attack / release times Different compressors don't give access to the same att/rel times. For instance, digital compressors propose much shorter attack times than optoelectronic ones. - attack / release curves Different technologies or brands bring different att/rel curves. An optoelectronic compressor, for instance, has very specific att/rel curves, which lead to a very specific grain. - distortion (1) When compressing, transients are very likely to bring distortion, since attack time must have a given length, and there is always a bit of transient remaining uncompressed. Thus, the kind of compression that happens inside the compressor, after the compression stage, is very important to the final result, since it will happen on each attack that gets compressed. - distortion (2) When using a compressor, it's common to feed the machine with a very high source level. Why ? It is true that I can always modify the compressor's threshold, telling it "above that limit, those are high levels that ought to be compressed ; below that limit, don't do anything". Apart from the fact that some machines have no threshold settings (rare), and input level is the only way to decide where the threshold is, high input levels take advantage of the compressor's first stage of amplification, which may have a distinctive timbre. In that case, the distortion coming from the machine's first stage (before the compression stage) is also very important.