Talk:Audio level compression - Definition and Overview

Note: Some material in this article was originally from Federal Standard 1037C, which was written in support of MIL-STD-188.

Additions and details

If the following information is correct, please add it to the article:

compression vs limiting

needs to show the difference between compression and soft clipping, which always confused me. i'm pretty sure i understand it now, but i bet others are confused too. compression will apply a scaling or attenuation to an entire wavelength when the peaks of that wave get above a certain level. soft limiting will apply an attenuation to only the parts of the wave that actually go above that level, reshaping the waveform. - Omegatron 17:36, Sep 1, 2004 (UTC)

companding

I believe companding is more of a non-linear effect like limiting. for instance, a signal can be exponentially companded which scales the entire waveform according to an exponential curve, so that the quiet parts are slightly less loud than the unchanged loud parts, it is passed over the noisy channel, and then the exponential non-linear function is done backwards, to make the quiet parts quiet again. i don't think it uses thresholds or changes gains anywhere. this, compression, limiting, and other forms of amplitude modification should be clarified. - Omegatron 17:45, Sep 1, 2004 (UTC)

http://www.arts.arizona.edu/studio/Compressors,LimitersExp.html

that site has different definitions for both, it seems. there is a lot of confusion about the specifics. i will try to learn the actual properties and draw some graphs. - Omegatron 17:52, Sep 1, 2004 (UTC)

http://www-ccrma.stanford.edu/~jos/filters/Dynamic_Range_Compression.html

yep. according to http://diana.ee.pucrs.br/~decastro/pdf/spra163a.pdf :

The U.S. and Japan use m-law companding. Limiting sample values to 13 magnitude bits, the m-law compression portion of this standard is defined mathematically by the continuous equation:

F(x) = sgn(x) ln(1 + μ |x|) / ln (1 + μ) Equation (1)

-1 ≤ x ≤ 1

where μ is the compression parameter (μ=255 for the U.S. and Japan), and x is the normalized integer to be compressed.

That's not the same as regular audio compression, which "reacts" to a loud signal by turning down the gain. Instead it's an equation that reshapes the signal to a logarithmic/exponential curve. These all need to be clarified. - Omegatron 16:18, Sep 2, 2004 (UTC)

Graphs

My impression is as follows:

Regular waveform:

Waveformoriginal.png
Image:Waveformoriginal.png

Compressed waveform:

Waveformcompressed.png
Image:Waveformcompressed.png

Hard limited or clipped waveform:

Waveformclipped.png
Image:Waveformclipped.png

Soft limited waveform:

Waveformsoftlimited.png
Image:Waveformsoftlimited.png

First stage of companding:

Waveformcompanded.png
Image:Waveformcompanded.png

it sort of maps all the amplitude values to an exponential curve (this example uses the original function^(1/5)), so the quieter parts are louder, but it is nonlinear distortion at the same time.

although this particular signal gets louder, so i guess it is expansion? kind of confusing.

no, it is compression, since it is compressing the dynamic range of the signal. however, it has a higher gain than the first example. the rule is that compression always decreases dynamic range and expansion increases it. - Omegatron 16:03, Sep 2, 2004 (UTC)

My field is communications rather than music so terminology could differ. I think that graphs showing waveforms are not good for representing compressor action. Compressors are linear; sine waves in will be sine waves out after level adjustment.

Linear on short time scales, you mean. The gain is just changed dynamically, right? The compressed waveform image is supposed to show that, with the compression acting halfway through... - Omegatron

The gain changes are much slower than the waveform.

Oh yeah. Good point. - Omegatron

Limiting is an ambiguous term, it could mean a sort of modified clipping as you are representing or a fast-acting high-level linear compressor. Meggar 22:18, 2004 Dec 8 (UTC)

Ok. I would still like some images in the article, but I don't know how to show them accurately. - Omegatron 22:29, Dec 8, 2004 (UTC)

Yeh these graphs are definatly not correct. The reaction speed of a compressor is at least one period of the slowest signal present (ie: the bass end) and usualy slower. A better graph would be showing loudness over time. For example the opening movement of Beethoven's 9th starts with a very loud burst of strings and tympany, followed by a much quieter bit. After compression the loud burst would be much quieter, and since the gain is usualy turned up to compensate for the compressor's action, the quieter bit would be louder. I've added an example of TV advertizing, where compression is used liberaly on the sound track to make it sound louder. --Swamp Ig 01:36, 21 Jan 2005 (UTC)

Thanks. I still want graphs, though. So I would use something like this kind of waveform: [1] (http://www.ee.columbia.edu/~marios/projects/dct_listening/waveform.png), and then have sections of the "before compressor" and "after compressor" blown up to show that the wave has not been distorted nonlinearly in the process? The limiting and soft limiting are correct, right? I will put them in the appropriate articles. Not sure about the compander waveform. - Omegatron 14:58, Jan 21, 2005 (UTC)

Here's a crude idea for an image. The inset is supposed to show the same waveform simply at a lower gain level. I'll fool around with audio programs later and maybe make it with those. Is this a good way to show the relationship? - Omegatron 22:24, Jan 21, 2005 (UTC)

more images

Images of limiting and clipping: