Dynamic range

[Buckle-meister]

All,

Is the volume at which music is recorded onto CD related to the dynamic range of the music or are they two separate issues? In other words, is it possible to record music onto CD quietly so that you have to turn the volume up reasonably high to hear the music even though the music itself is regarded as 'loud' from having poor dynamic range?

 

[MDS]

Dynamic range and the average level of the music are related but not perfectly so. First we have to clarify some of the terms because 'dynamic range' is often used as a catch-all term when something slightly different is actually meant.

Dynamic Range
Generally means the difference between the softest parts and the loudest parts, expressed in decibels. Even on some current CDs that are 'loud' you will see the peak level meters (using Sound Forge, for example) drop way down to -50 dB during some parts of the music even though the music peaks at 0 dB. That would mean the dynamic range is roughly 50 dB.

Average RMS Level
Average Level is what is referred to when it is said that the CD is recorded 'hot'. It expresses how far from 0 dB (max) is the average of all the sample values. Current CDs (from my collection at least) have average levels of -10 dB to -12 dB and peak at 0 dB. Older '80s CDs have average levels of -16 dB to -18 dB and except for a few moments rarely ever peak at 0 dB.

Crest Factor
Crest Factor is really what people mean when they say the CD only has '6 dB of dynamic range'. It is the difference between the Average Level and the Peak Level. It is not as simple as just subtracting the average from the peak though - the level may average -10 dB but much of the waveform may spend it's time between -6 dB and the peak of 0 dB.

Perceived Loudness
The level of 'loundess' that you perceive is related to the Average Level but the content of the music influences your perception more than the average level. Ozzy Osbourne at -18 dB average level is still going to be perceived as louder than Barry Manilow at -18 dB.

Most of the increase in loudness levels comes from use of dynamic compression and limiters. The waveform is squashed, bringing up the average level. But a waveform with a relatively high average level can still exhibit good dynamic range and a crest factor that is higher than what you often see on newer CDs if only certain portions of the waveform are compressed.

If you open a waveform in Sound Forge and it looks like a square when viewed at the default zoom ratio of 4096:1, it will almost certainly have a high average level and poor crest factor. Others have fairly high average levels but lots of spacing between the peaks and sound just fine.

 

[The Dukester]

Ozzy Osbourne at -18 dB average level is still going to be perceived as louder than Barry Manilow at -18 dB.

Common, MDS. Barry Manilow at -100 is loud.

Nice post.

 

[Buckle-meister]

...on some current CDs that are 'loud' you will see the peak level meters...drop way down to -50 dB during some parts of the music even though the music peaks at 0 dB. That would mean the dynamic range is roughly 50 dB.

Whilst it's bad enough that music is compressed at all, if it's going to be, why not place it on the CD so that the music's peak level lies beneath 0dB? CDs have a theoretical dynamic range of 96dB don't they? Then why, in the above quote, not place the (compressed) music so that its dynamic range falls from, say, -60dB to -10dB?

Average Level is what is referred to when it is said that the CD is recorded 'hot'. It expresses how far from 0 dB (max) is the average of all the sample values. Current CDs (from my collection at least) have average levels of -10 dB to -12 dB and peak at 0 dB. Older '80s CDs have average levels of -16 dB to -18 dB and except for a few moments rarely ever peak at 0 dB.

So, content with an average level greater than half the dynamic range (assuming a peak level of 0dB) says nothing other than that the music has fewer lulls than loud passages. In other words, you cannot determine from average level whether or not the music's been compressed. Correct?

 

[MDS]

Whilst it's bad enough that music is compressed at all, if it's going to be, why not place it on the CD so that the music's peak level lies beneath 0dB? CDs have a theoretical dynamic range of 96dB don't they? Then why, in the above quote, not place the (compressed) music so that its dynamic range falls from, say, -60dB to -10dB?

The range would still be the same 50 dB.

If you have a waveform that is very compressed, with lots of consecutive peaks at 0 dB, it will light the clip indicators. My experience is with Sound Forge and it defines clipping as 4 consecutive 0 dB peaks. If you take that waveform and normalize it down -10 dB, now the peak level will be -10 dB and the clip indicators will never light - but you haven't changed anything - the waveform is still compressed and just as loud but it is smaller.

If you look at my Waveforms Illustrated thread, I picked Oasis - Hello to illustrate that case. That peaks at -3 dB but is incredibly compressed and has a high average level.

So, content with an average level greater than half the dynamic range (assuming a peak level of 0dB) says nothing other than that the music has fewer lulls than loud passages. In other words, you cannot determine from average level whether or not the music's been compressed. Correct?

Compressors and Limiters are just tools of the trade and are not inherently bad. Neither is peaking at 0 dB. The problem is with excessive compression where there are thousands of peaks at 0 dB as opposed to the way it used to be where only a few select passages would peak that high. In the case where the average level is high and it constantly peaks at 0 dB there are fewer lulls than loud passages as you've said. It's the 'everything is loud' part that is the problem.

 

[Buckle-meister]

The range would still be the same 50 dB.

Yes, I understand that, but I figured that if you have options of:

1. Compressing music and placing it on a CD such that the peak level is at 0dB or;
2. Compressing music but placing it on a CD such that the peak level is below 0dB...

...option 2 would be better, or rather, the lesser of two evils. No?

Compressors and Limiters are just tools of the trade and are not inherently bad. Neither is peaking at 0 dB...

Why isn't peaking at 0dB bad? Especially when, I would've thought, virtually all music with the exception of orchestral (which can exhibit large level differences throughout a piece) can easily 'fit' within CD's 96dB range.

In other words, why, if a piece of music has a dynamic range (compressed or uncompressed) of say, 60dB which, I gather, is pretty rare, isn't it placed from -96 to -36dB or -90 to -30, but not -60 to 0dB?! Why shunt the 60dB 'block' right to the very limit of the disc when there's no need? Or is there and I'm just missing something?

If you look at my Waveforms Illustrated thread...

Just going there now.

 

[MDS]

When the music is recorded, the levels are set so that the peaks don't hit 0 dB. It would be like the old days of recording to tape where you wanted to set the levels so that the VU meter only occasionally goes into the red. Unless you listen to the whole song from start to end you can't tell just how high the peaks will go so room is left to ensure that you don't clip.

It's the post-processing after you have the raw track where the compressor and limiter are used. If you had recorded so that the peaks are about -3 dB you have 3 dB of headroom to use to make it louder. But if you have only a few peaks that hit -3 dB then an operation like normalizing to 0 dB will just add 3 dB to every sample. It will only be slightly louder. With compression you can bring up the level over the entire song.

Maybe tonight I'll post an example of light pop music like Air Supply and show the original with a low average level but a few peaks near -3 dB, the result of normalizing that to 0 dB, and the result of compressing the same thing to bring the average level way up.

 

[tbewick]

Yes, I understand that, but I figured that if you have options of:

1. Compressing music and placing it on a CD such that the peak level is at 0dB or;
2. Compressing music but placing it on a CD such that the peak level is below 0dB...

...option 2 would be better, or rather, the lesser of two evils. No?

I think it's a case of getting the best signal-to-noise ratio, that is by having the music normalised to extend across the full possible range of the digital system. For instance, I have used a sound editor and it gives you the option to normalise to 100%, which means the maximum signal level is made 0 dB full-scale.

There is an argument that recordings should not be normalised to full-scale, because digital converters may overload near 0 dB full-scale. This would be most problematic with heavily compressed material, since the signal intensity is near full-scale very often. Thomas Lund has discussed this issue at a recent AES conference:

http://www.tcelectronic.com/Default.asp?Id=9249&AjrThmPg=0

Why isn't peaking at 0dB bad? Especially when, I would've thought, virtually all music with the exception of orchestral (which can exhibit large level differences throughout a piece) can easily 'fit' within CD's 96dB range.

In other words, why, if a piece of music has a dynamic range (compressed or uncompressed) of say, 60dB which, I gather, is pretty rare, isn't it placed from -96 to -36dB or -90 to -30, but not -60 to 0dB?! Why shunt the 60dB 'block' right to the very limit of the disc when there's no need? Or is there and I'm just missing something?

I think the primary reason for signals being normalised to 0 dB full-scale is to get the best possible performance from the available resolution of the digital system. The problem is, however, that practical digital converters do not function as well as the theoretical level of performance available from a given number of bits, i.e. they may show non-linearity near full-scale.

I can understand why the practice of mastering to full-scale has been done, since no books I've read on the subject mention the practical issues of converter performance near full-scale. I think the reason for this is that converter designers maybe work to the assumption that headroom will be left near 0 dB full-scale. This is how sound level meters work, where -12 dB full-scale corresponds to peak level on the analog meter (or -18 dB full-scale in Europe). I believe that after this, the indicator will be in the yellow and red (at 0 dB full-scale) regions. Some engineers may incorrectly assume that a digital system will show no non-linearity all the way up to 0 dB full-scale, so compress their recordings to this level. It wouldn't matter for the occasional peak to go into digital clipping, but highly compressed recordings often spend all their time in this region.