[limpidglitch]

When you guys talk about music being more or less 'dynamic', what are you really talking about?
Is there a clear definition? Can it be quantified?

 

[pinnahertz]

  When you guys talk about music being more or less 'dynamic', what are you really talking about?
Is there a clear definition? Can it be quantified?

I refer to Dynamic Range.  I like the definition that pops first in Google:
 
dy·nam·ic range
noun
 

• the range of acceptable or possible volumes of sound occurring in the course of a piece of music or a performance.
  
  1. the ratio of the largest to the smallest intensity of sound that can be reliably transmitted or reproduced by a particular sound system, measured in decibels.
     
     
  
  
  
  

 
 
It can be quantified a number of ways, but one easy one is the TT Dynamic Range Meter.
 
So if a track is "more dynamic" or has "greater dynamics", it would have greater dynamic range.  Dynamic range is very commonly reduced with dynamics processing, generally like compression and limiting, but usually more complex than that.  When dynamic range is reduced, the average loudness can be pushed up.  The competitive application of this kind of processing is what the "loudness war" is all about.  Modern EDM generally has very little dynamic range, especially when compared with other genres like classical. 

 

[StanD]

  I dunno...retro is in!

I can't wait to hear retro-hip-hop. I might have to get a retro headphone cable to listen to it.

 

[pinnahertz]

 I dunno...retro is in!

I can't wait to hear retro-hip-hop. I might have to get a retro headphone cable to listen to it.

Yeah, well, retro is a nice word for "old". You know, when Hip-hop become a hip replacement. Etc.

 

[StanD]

Yeah, well, retro is a nice word for "old". You know, when Hip-hop become a hip replacement. Etc.

About now, Hippies are ready for hip replacements and audiophile hearing aids to replace their Koss Pro4A headphones.

 

[gregorio]

  In my experience, edm tends to be annoyingly bright.

 
Agreed but "bright" and ultrasonic are not the same thing. "Bright" can inhabit quite a wide range of freqs, anything from about 2kHz up to 10kHz or so but typically around 4kHz to 7kHz. As the loudness wars intensified, EQ use often became more extreme in mastering in EDM, with particular emphasis on the low and the mid/high mid freqs.
 

Not so contemporary with almost same DR as Stockhausen you may try Jean Michel Jarre 80's production ( Oygen/Magnetic Fields/China's concert).
I am afraid it is old fashioned EDM

 
The history of EDM is quite colourful, it's roots can be traced back to Stockhausen in the '50's,  then in the early '70's came Kraftwerk who were arguably the single biggest influence. Neither Stockhausen (obviously), Kraftwerk or Jean Michel Jarre's work was strictly EDM though, just EDM's ancestors. Although admittedly EDM is not a very precisely defined term.
 

  When you guys talk about music being more or less 'dynamic', what are you really talking about?
Is there a clear definition? Can it be quantified?

 
Good question! Dynamics/dynamic range are not precisely defined terms. This is partly due to the fact that it's used as a musical term (by musicians) and as an audio engineering term.
 
In music, it's typically used to mean the range of loudness and/or the speed of change. A "very dynamic" composition would be one which relatively often changes between piano and forte (or pianissimo and fortissimo) and those changes are relatively abrupt, making the contrast between quiet and loud particularly obvious. In music, "dynamic/s" is more about the amount of this perceived contrast rather than absolute difference. For example, the opening of Beethoven's 5th Symph was shockingly "dynamic" in it's day, due to it's instant change from near silent to forte.
 
Even if we just take the audio engineering (rather than the musical) use of the term, dynamic range is still not precisely defined or more accurately, it is quite precisely defined but there's more than one such definition. Pinnahertz's quoted definition is perfectly accurate and acceptable but arguably, a more commonly used definition would be the range from peak value to the noise floor.
 
To simplify; the difference between the audio engineering definition/s and the musical definition is that the engineering definition/s are an actual (amplitude) numerical value whereas the musical definition is perception based. This is an important difference because particularly over the last 30 years, popular music genres have evolved to take advantage of it, EDM arguably more than any other. In other words, EDM (and other popular genres) can often appear reasonably dynamic but is in reality much less so than it appears. This is because song construction (the musical structure, arrangement/orchestration and processing) has evolved to take advantage of those factors which affect our perception of loudness.
 
G

 

[limpidglitch]

 
Even if we just take the audio engineering (rather than the musical) use of the term, dynamic range is still not precisely defined or more accurately, it is quite precisely defined but there's more than one such definition. Pinnahertz's quoted definition is perfectly accurate and acceptable but arguably, a more commonly used definition would be the range from peak value to the noise floor.
 
To simplify; the difference between the audio engineering definition/s and the musical definition is that the engineering definition/s are an actual (amplitude) numerical value whereas the musical definition is perception based. This is an important difference because particularly over the last 30 years, popular music genres have evolved to take advantage of it, EDM arguably more than any other. In other words, EDM (and other popular genres) can often appear reasonably dynamic but is in reality much less so than it appears. This is because song construction (the musical structure, arrangement/orchestration and processing) has evolved to take advantage of those factors which affect our perception of loudness.
 
G

 
Yes, Pinnahertz' definition sounds reasonable when applied to something like the transmission of signals over radio, or recording of light intensities on a CCD, but how does it relate to recorded music? 
To define a range you'd need to define the extremes, so how do you determine what's the highest and lowest levels?

I've also tried to figure out what the TT meter actually measures, but documentation seems hard to find. I believe RRod has mentioned that it's more or less a variation on crest factor, but crest factor doesn't describe a range, as far as I can understand, but the ratio between the peak sample value and the total RMS of the track.

 

[pinnahertz]

   
Yes, Pinnahertz' definition sounds reasonable when applied to something like the transmission of signals over radio, or recording of light intensities on a CCD, but how does it relate to recorded music? 
To define a range you'd need to define the extremes, so how do you determine what's the highest and lowest levels?

I've also tried to figure out what the TT meter actually measures, but documentation seems hard to find. I believe RRod has mentioned that it's more or less a variation on crest factor, but crest factor doesn't describe a range, as far as I can understand, but the ratio between the peak sample value and the total RMS of the track.

The TT DR meter is an attempt to reduce a complex metering process down to a single digit for publication.  It's derived from taking the difference between RMS and Peak, but time weighting so it makes more sense.  As far as I can tell the peak figure is derived from a true peak detector with defined ballistic.  This would be logical as time is also required to perceive loudness.  
 
However, the meter relates to short-term DR, not the total DR of a piece of music.  For example, a recording of the last movement of Beethoven's 9th returns a DR 13, but the total dynamic range per the first definition I posted is in the 40dB area.  However, a widely dynamic piece like Beethoven's 9th will always return a higher number than a EDM piece, so it does serve to illustrate the principle, even if its massive integration is a bit misleading when considering total DR.  The TT meter also relates well with loudness meters, even though the numbers are different.  The point was to present an idea of the degree of loudness processing in a piece, though it's not a loudness meter as such, nor will it display total DR.  
 
I only suggested the TT meter because it's quick and easy.  I use the tools in Audition mostly, because it presents lots of different types of analysis and can span part of a file or the whole thing. I also us an Leq analyzer.  

 

[limpidglitch]

What do you mean with "time weighting"? Do you know of anywhere I can read about the algorithm in detail?
 
I've also used Audition a lot.
Their 'Dynamic Range' is the difference between the maximum and minimum RMS values, which introduces another complication: what window length should you use? If you set it short enough it wouldn't be difficult to make Skrillex and Beethoven appear similarly dynamic.

 

[gregorio]

  [1] Yes, Pinnahertz' definition sounds reasonable when applied to something like the transmission of signals over radio, or recording of light intensities on a CCD, but how does it relate to recorded music? 
[2] To define a range you'd need to define the extremes, so how do you determine what's the highest and lowest levels?

[3] I've also tried to figure out what the TT meter actually measures, but documentation seems hard to find. I believe RRod has mentioned that it's more or less a variation on crest factor, but crest factor doesn't describe a range, as far as I can understand, but the ratio between the peak sample value and the total RMS of the track.

 
1. Well it would relate more to sound systems or components than to recorded music, as you would ideally want a sound system's dynamic range to be greater than the dynamic range of the recordings. IE. So that it can in effect reproduce the recording's noise floor.
 
2. The highest level is easy because we can just assume it's 0dBFS, as pretty much all commercial music recordings peak within a few tenths of a dB of 0dBFS. The lower figure is not so easy in practice because although in theory it's the recording's noise floor, in practice the noise floor typically isn't completely constant and depending on how random the noise floor, it's spectrum and the spectrum of the signal, it is possible to hear notes several dB below the noise floor. In extreme examples (such as noise-shaped dither for example) it can be possible to hear 30dB below the noise floor.
 
3. AFAIK, the actual measurement/algo is not published. It is roughly, as RRod describes a measurement of crest factor and I also believe Pinnahertz is correct in stating it effectively has a time "window", possibly similar to the old PPMs. Whatever it is actually doing, it's really nothing to do with actual dynamic range though.
 
G

 

[pinnahertz]

  What do you mean with "time weighting"? Do you know of anywhere I can read about the algorithm in detail?
 
I've also used Audition a lot.
Their 'Dynamic Range' is the difference between the maximum and minimum RMS values, which introduces another complication: what window length should you use? If you set it short enough it wouldn't be difficult to make Skrillex and Beethoven appear similarly dynamic.

True peak is instantaneous.  Any peak of any duration is detected at its maximum value.  What I mean by "time-weighting" is a peak detector that will eventually respond to true peak, but only after a period of time.  What that does is weight the reading toward longer duration peaks, and read very short peaks at a lower that true value, corresponding to how we perceive the build up of sound.  
 
For the analysis window in Audition, you have to be a bit intelle\igent about what you select.  Longer is better, generally, but any time you pick a portion of a track you're biasing the reading somewhat. I always start by analyzing the entire track, then go in for detail if required.  Analyzing the total track will give you the maximum and minimums for peak, RMS, and average, and show you all possibly clipped samples over 0dBFS, total track loudness (including ITU-R BS.1770-2).  Doing less than the whole track creates both specificity (analyzing just one specific portion) and ambiguity (the numbers say nothing about what section it is you've analyzed). 

 

[RRod]

I find the DR rating to be mostly useful for comparing different versions of the same track. The dynamic structure of two pieces can easily vary enough that the rating is no longer measuring a "difference in dynamism" but a "difference in suitability for the algorithm". Some material just doesn't work well with it: take Jon Leifs' music as an example. He has several big works that are essentially really long crescendos. This means that when you only consider, say, some quantile range of RMS values, you are cutting off the loudest and the softest parts! Bad way to get a true measure of dynamism. The more I've worked with this stuff the more I think what G suggests is what works: a "peak-ish" to noise-floor measurement. I say "peak-ish" because once you get to really short durations, the integration abilities of our hearing come into play.

 

[CarlosUnchained]

I really like this thread. Free audio engineering lessons.

 

[limpidglitch]

No Skrillex at hand, so it's Kiasmos vs. some classy Ludvig van.
I'd say Ludvig sounds more dynamic, but at the same time you could argue Kiasmos has more actual dynamic range.
 

The tracks were 65 by Kiasmos and 9th Symphony, 4th movt. by LvB/Fricsay/Berliner Phil.
The tracks were summed to mono and trimmed 10 seconds on either end before processing.
 
 

Spoiler: SoX/Gnuplot code

#compare dynamic range* of two files
#requires SoX and Gnuplot
#*(here defined as difference between minimum and maximum dB RMS)
 
#sum to mono, trim first and last 10 seconds and normalize the audio files
sox DR_Kiasmos-65.wav audio1.wav channels 1 trim 10 -10 norm -0.001
sox DR_LvB.Fricsay-s9.4.wav audio2.wav channels 1 trim 10 -10 norm -0.001
 
#initiate data file with comment and headers
printf "#max-min RMS range\nSec Kiasmos LvB\n" > DR.txt
 
#repeat calculations with increasing window lengths
for i in {1..100}
do
    width=$(echo "scale=3; $i / 100" | bc -l | sed 's/^\./0./g')
    echo "$width"
 
    rmsmax1=$(sox audio1.wav -n stats -w $width 2>&1 |\
      grep "RMS Pk dB" |\
      sed 's/[^0-9.-]*//g')
    rmsmin1=$(sox audio1.wav -n stats -w $width 2>&1 |\
      grep "RMS Tr dB" |\
      sed 's/[^0-9.-]*//g')
    rmsdif1=$(echo "($rmsmax1)-($rmsmin1)" | bc -l | sed 's/^\./0./g')
 
    rmsmax2=$(sox audio2.wav -n stats -w $width 2>&1 |\
      grep "RMS Pk dB" |\
      sed 's/[^0-9.-]*//g')
    rmsmin2=$(sox audio2.wav -n stats -w $width 2>&1 |\
      grep "RMS Tr dB" |\
      sed "s/[^0-9.-]*//g")
    rmsdif2=$(echo "($rmsmax2)-($rmsmin2)" | bc -l | sed "s/^\./0./g")
 
    echo "$width $rmsdif1 $rmsdif2" >> DR.txt
done
 
cat DR.txt
 
gnuplot
 
set term png size 2000,1000 font ",20"
set output "DR.png"
 
set xlabel "Window length (s)"
set ylabel "Range (dB RMS)"
 
set key height 1
plot for[col=2:3] "DR.txt" using 1:col lw 3 title columnheader(col) with lines
 
exit
 
open DR.png
 

 

[pinnahertz]

No Skrillex at hand, so it's Kiasmos vs. some classy Ludvig van.
I'd say Ludvig sounds more dynamic, but at the same time you could argue Kiasmos has more actual dynamic range.
 

The tracks were 65 by Kiasmos and 9th Symphony, 4th movt. by LvB/Fricsay/Berliner Phil.
The tracks were summed to mono and trimmed 10 seconds on either end before processing.
 
 

Spoiler: SoX/Gnuplot code

#compare dynamic range* of two files
#requires SoX and Gnuplot
#*(here defined as difference between minimum and maximum dB RMS)
 
#sum to mono, trim first and last 10 seconds and normalize the audio files
sox DR_Kiasmos-65.wav audio1.wav channels 1 trim 10 -10 norm -0.001
sox DR_LvB.Fricsay-s9.4.wav audio2.wav channels 1 trim 10 -10 norm -0.001
 
#initiate data file with comment and headers
printf "#max-min RMS range\nSec Kiasmos LvB\n" > DR.txt
 
#repeat calculations with increasing window lengths
for i in {1..100}
do
    width=$(echo "scale=3; $i / 100" | bc -l | sed 's/^\./0./g')
    echo "$width"
 
    rmsmax1=$(sox audio1.wav -n stats -w $width 2>&1 |\
      grep "RMS Pk dB" |\
      sed 's/[^0-9.-]*//g')
    rmsmin1=$(sox audio1.wav -n stats -w $width 2>&1 |\
      grep "RMS Tr dB" |\
      sed 's/[^0-9.-]*//g')
    rmsdif1=$(echo "($rmsmax1)-($rmsmin1)" | bc -l | sed 's/^\./0./g')
 
    rmsmax2=$(sox audio2.wav -n stats -w $width 2>&1 |\
      grep "RMS Pk dB" |\
      sed 's/[^0-9.-]*//g')
    rmsmin2=$(sox audio2.wav -n stats -w $width 2>&1 |\
      grep "RMS Tr dB" |\
      sed "s/[^0-9.-]*//g")
    rmsdif2=$(echo "($rmsmax2)-($rmsmin2)" | bc -l | sed "s/^\./0./g")
 
    echo "$width $rmsdif1 $rmsdif2" >> DR.txt
done
 
cat DR.txt
 
gnuplot
 
set term png size 2000,1000 font ",20"
set output "DR.png"
 
set xlabel "Window length (s)"
set ylabel "Range (dB RMS)"
 
set key height 1
plot for[col=2:3] "DR.txt" using 1:col lw 3 title columnheader(col) with lines
 
exit
 
open DR.png
 

The graph shows the analysis window at 1 sec.? That's not right....