Dynamic Compression

There has been a battle with record companies to produce music that is louder than their competitors. There is no good reason why we should buy music that has been processed to the point where it is painful to listen to. That is, unless you enjoy pain.

Note that in this case, “compression” refers to an amplification process; not the file compression that occurs when you save music in a compressed file format like MP3 or AAC. That is another topic altogether.

Let me show you what I’m talking about. Here is a visual representation and an audio sample of a piece of music produced in the early 1990s.

Figure 1. Wave form display of a short mix of Toto’s “Rosanna”.
Sound sample 1: Sound sample of Figure 1.

Today, if this was offered as “remastered” or re-released in a compilation album, chances are the wave would look and sound something like this:

Figure 2. Wave form display of a short mix of Toto’s “Rosanna”, dynamically processed, limited and normalized.
Sound sample 2: Sound sample of Figure 2, volume reduced for comparison purposes.

The second sound sample has an overall loudness compared to the first sample, and it resembles audio which you may hear from FM radio (due to radio’s automatic gain control [AGC] limiting).

Why record companies would want to take clean music and make it sound like it’s being sent through FM radio is beyond my comprehension.

Here is how it is done (or I should say how it shouldn’t be done).

One of the first steps to maximize volume is to apply a dynamics filter. This increases the quiet sections while at the same time, reducing the loud sections. Figure 3 shows that some amount of loudness has been applied to the original wave. Compare Figure 3 to Figure 1.

This is a graph showing input levels on the x axis and output levels on the y axis. Levels that are -30 dB will be brought up to about -23 dB while levels at 0 dB will be brought down to -15 dB. Everything else is defined along the curve.

Figure 3. The wave after applying a dynamics filter.

The next step is to apply a limiter. In this case, about -10 dB, and squishes anything outside that level to fit within that range. It’s important to note that this process is not the same as simply cutting off everything above -10 dB. That wound result in clipping, which is a deadly sin in digital audio.

Without this step, the peaks in Figure 3, although faint in the image, are large enough to affect the normalization process.

Figure 4. Limited at -10 dB, ready to normalize.

One final step is to apply normalization, which finds the maximum peak in the file, which we know is -10dB because of the previous step, and amplifies the entire sound by a constant factor so that the maximum peak is the target amplitude, in this case, -0.1 dB.

This results in the final compressed wave as shown in Figure 2.

If you purchase music that has been processed in this way, I would personally ask for your money back. Somewhere along the way, there is a much brighter version of the song before it was destroyed in this process. By the time the song gets into your hands or in your ears, there is no way you can ever recover its original wide dynamic range.

My personal feeling on this matter is that home or car stereo systems, or even MP3 players, should have options built in so that you can choose to dynamically process (as radio stations use automatic gain control [AGC] to maximize the volume of their programming material) music depending on your environment. While driving, you probably want to maximize the volume to compensate for the environmental noise, but in the quiet of your home on your high-fidelity sound system, you most likely want the full dynamic range.

More Information

Loudness Wars

Dynamic Range Database

Article History

2006-10-15: Initial posting

2009-07-19: Updated music clips

2017-09-03: Updated music clips, changed “CD” to “digital audio”