This note is intended to clarify the sometimes confusing terminology
This is the level, measured in dB below peak, below which no alteration of the audio will take place. If the threshold is -1dB then practically no compression will take place. If it is -40dB then pretty well everything will be compressed.
Above the threshold any rise in peak level will cause the level of the audio to be reduced. With a 2:1 ratio a 2db increase will be reduced to become a 1 db increase and pro rata. With a 5:1 ratio a 10db increase will be reduced to be a 2 dB increase.
Since the compression reduces the level of the louder parts of the track it is normally necessary to increase the gain in order to bring the peaks back to where they were. This of course results in the quiet (uncompressed) sections being brought up. This is usually applied after the compression. If it is applied before it will be as if you had applied it afterwards but then reduced the threshold.
If the audio is quiet and suddenly goes loud the compression can kick in so fast that the audio is ducked instantaneously, or can be slow so that the level will be reduced more gradually - this will let through sudden peaks. A low attack time is usual - a few ms (milliseconds - thousandths of a second).
RELEASE TIME (RECOVERY TIME)
If the audio has been loud and suddenly goes quiet so that compression is no longer taking place, then the level of this quiet section will be raised back to where it would be with no compression. This can be fast - which can be audible as it comes up suddenly and causes 'pumping' - or slow, so that the recovery is gradual.
Sophisticated compressors can behave differently over different frequency bands.
To get a visual idea of compression, look at this picture of a compressor in Cubase LE:
Look at the square graph in the top left-hand corner, labelled 'Characteristic'. The horizontal axis represents the level of the incoming audio: the vertical axis represents the output after compression.
Note that the line is at 45 degrees from low to -20: this means that up to -20dB the output matches the input. You will see a 'knee' in the graph at -20 and above this the line is flattened: at about -20 input the output is -10 and rises to -1.5 as the input rises to peak - a compression ratio of 2:1 with make-up gain of 8.5 dB.
A limiter is in effect a compressor with an infinity:1 ratio, so that above a specified threshold (usually maximum peak) there is no further gain no matter how much extra level you put in: it's nasty in effect but useful for preventing sudden accidental over-peaks.
Note that we are talking about compression of the dynamic (loud-to-soft) range of audio, not to be confused with digital compression as used in making MP3 files. This is entirely different and works by leaving out bits of the audio which it is calculated the ear will hardly miss.
Compressor-expanders are a special case (not relevant to normal audio processing) designed to compress the audio before recording and expand it on playback so as to minimize tape hiss. The simplest system is DBX, which applies 2:1 compression over most of the dynamic range: unfortunately this method can cause the hiss to be heard 'pumping'. The best known is Dolby B, ubiquitous on cassette recorders, which affects only the higher frequencies: unlike a conventional compressor it affects only audio below a certain threshold, compressing as the audio drops below this until there has been about a 10dB increase in the quietest levels. (The actual curve used is a little more complicated than a straight compressor). On playback the process is exactly reversed (requiring accurate level alignment to get the threshold in the right place) and reduces tape hiss by 10dB with remarkable effectiveness, having practicably no detectable effect on the actual audio. Later developments in the professional field - Dolby A and Dolby Spectral - work in a more complex fashion with more frequency bands but the basic principle is the same. Dolby C is the domestic version of Dolby Spectral and is remarkably effective on cassette recordings. However the entire technique has largely been rendered obsolete by digital recording.
© Roger Wilmut