I keep mentioning that linear compression doesn’t work very well. So maybe it is about time I show you why…
This graph shows what happens, in a frequency channel exhibiting 60 dB of threshold elevation, when I try to squeeze the dynamic range of my music into 1/3 of its original range, using a linear compressor with ratio 3:1.
There’s a lot going on in this graph. The black diagonal line is our reference ideal. That’s what people without hearing impairment would hear – whatever comes in is what they hear. That’s also what a person using a Crescendo would hear, even if they had this hearing impairment.
In order to reach that ideal of hearing, we have to use nonlinear compression, whose compression curve is shown in the light magenta curve. In comparison, our linear compression curve is shown in light orange. It has a slope of 1/3, with a threshold 24 dB below our nominal 0 dBVU = 77 dB. Below that threshold it simply supplies a constant gain.
And you can see how it compares to a proper compression curve, and how it really is correct, but only at two loudness levels, where it intersects the nonlinear compression curve.
Now the red curve is uncorrected recruitment hearing for this threshold elevation. The bold green curve shows the hearing perception after using the linear compressor.
Linear compression does help – of course! Anything helps! But it isn’t correct. It produces dynamic range reduction. It does allow us to hear below our previous limitations.
But now, you say, that isn’t really so bad… And I might agree if the application were only attempting to recover speech perception. Nobody really cares about fidelity when trying to understand someone speaking. That’s why we tolerate telephones.
But our hearing exhibits a changing degree of impairment with frequency. Higher frequencies show greater impairment. Compare this same compression scenario for a lower frequency band where the threshold elevation is only 40 dB:
Now you see that the degree of overcorrection is even greater than before, by a lot.
And in the region below 1 kHz there would be no need of compression because your hearing is likely close enough to normal.
So now, consider an oboe playing C5, an octave above middle-C. That’s about 500 Hz. (within Astronomical accuracy…) Its fundamental will not receive any correction. Its first and second harmonics at 1 kHz and 1.5 kHz will receive the treatment you see in this second graph. And its higher harmonics will get the treatment from the first graph.
All of a sudden we have exaggerated the first and second harmonics and the poor oboe ends up sounding very similar to a muted jazz trumpet.
That’s why using linear compressors, to squeeze dynamic range into narrowed perceptual ranges of damaged hearing, is completely horrible for music. We need something a bit more elaborate in order to preserve fidelity and not distort the harmonic ratios and musical timbre of instruments.
[ … and once again, the 1 kHz Cliff… Gross exaggeration of the “Pain Frequencies”, and that’s likely why we all hate the sound in our hearing aids… ]
[ Just as a quick experiment, I took a sample of an A440 oboe and a muted trumped, and played around with a graphics equalizer on the oboe until its spectrum more or less matched that from the trumpet. It took about 20 dB of boost in the 1-2 kHz region. Looking up at the second graph above, that’s just about what my equations would predict in the 55 dB range. Interesting… ]