General question on dynamic range and music

[Newman]

If you allow silences into the DR measurement, then you are not measuring an attribute of the music.

 

[Jochen]

No there is no necessity of correlation. Maybe from a wax cylinder where the surface noise is so high, but in modern tape based or digital recording other influences set the noise floor and it is well below the crest factor when music is playing.

Sorry to ask, do you know what a correlation coefficient is? To be specific, Pearson correlation?

 

[Jochen]

If you allow silences into the DR measurement, then you are not measuring an attribute of the music.

By the way, who says that this is silence ? It is very quiet, exactly because of this this recording has a high DR.

 

[Jochen]

The building up at around second 3 or 4 is definitely not silence. So even if you cut the measurement region before that and after around second 58, the result will not change much.

 

[Blumlein 88]

Sorry to ask, do you know what a correlation coefficient is? To be specific, Pearson correlation?

Yes.

 

[Jochen]

Yes.

Ok, good. I mean there may be exceptions, but for the most recording in whatever style, if the ratio of the maximum sound to the rms is high, so will be the ratio of the maximum to the lowest and vice versa. Thus high correlation. Don't you agree with that? Take a loundess war recording. The rms is pretty high and the maximum only slightly above, thus low crest factor. But the the lowest sound is as well only slighly below the rms, thus low dynamic range.

 

[Newman]

The building up at around second 3 or 4 is definitely not silence. So even if you cut the measurement region before that and after around second 58, the result will not change much.

It doesn’t matter which bits are not silence, it matters that some are. And by 'silence' I mean no music.

 

[krabapple]

Not red enough . . .

View attachment 283564

Van Halen's "Hot for Teacher" from the 2000 Warner HDCD remaster.

Here's Van Halen's "Hot for Teacher" from the 1984 album, 1983 Warner Bros. Records – 9 23985-2 Japan Target:

View attachment 283574

Much better . . .

FWIW the 2015 remaster of 1984 offered on HDTracks has full dynamic range....

...but only if you purchase the 192/24 version!

 

[Blumlein 88]

Ok, good. I mean there may be exceptions, but for the most recording in whatever style, if the ratio of the maximum sound to the rms is high, so will be the ratio of the maximum to the lowest and vice versa. Thus high correlation. Don't you agree with that? Take a loundess war recording. The rms is pretty high and the maximum only slightly above, thus low crest factor. But the the lowest sound is as well only slighly below the rms, thus low dynamic range.

Probably not. What might create some correlation is people who care about higher fidelity are less likely to crush things with over-compression. They also are likely to have a wider technical dynamic range. Yet, it is quite possible to create a low DR recording with the highest dynamic range gear. One is not a requirement or directly linked to the other.

I'll introduce some simple complications (is that an oxymoron?). Many people have listening rooms that on a sound level meter would read around 40 db SPL for noise. You'll find it recommended that your gear should have perhaps 105 db SPL capability in smaller listening spaces. Does that mean 65 db dynamic range is all you could possibly hear?

No. For one reason we can hear into noise perhaps 15-20 db. Yet that varies with frequency. So is 85 db dynamic range the most we could hear? No, because noise has a pinkish to brownish quality. Most of the energy is at low frequencies we don't have much sensitivity to hearing. Our hearing kind of breaks sound into ERB filters (effective rectangular bandwidth). In the band with the greatest sensitivity around 3 or 4 khz a room with 40 dbSPL noise might only have noise of 10 dbSPL or less. That is also the area we can hear most deeply into noise. So in such a somewhat noisy environment in some parts of the spectrum we might hear right down to near 0 dbSPL. So you might need gear that can do 105 db of dynamic range. OTOH, if someone hits you with 105 db even for a moment during a transient it will be some period of time before you could hear the lowest level signals even if the gear and recording has them.

Now further, one can create synthetic music with 140 db of dynamic range. If however the source is from a microphone, guitar amp or such there are limitations on how quiet it can be from background noise, microphone self noise, and noise in the microphone preamp. While one might process it to create more it is very uncommon to manage something with more than 80 db of dynamic range. Further for most music most people don't want that. They prefer a bit of compression. You might say you don't, and there are times I don't. Music being a commercial venture means they will cater to what most people want or use. Also surprisingly, some tastefully done compression will be perceived as much more dynamic than uncompressed music which has the far higher measured dynamic range.

 

[krabapple]

Nothing to do is probably not true. If the crest factor is high, that means that either the loudest signal is high or rms of the signal is low or both. In any case this will most probably also lead to a large "dynamical range" in the definition of the "softest (or noise floor) versus loudest recorded levels (the technical dynamic range)". Thus I would assume a quite high correlation between these two measures.

Crest factor and the "DR meter' reading can be affected by EQ (e.g. lots of bass in mastering A compared to mastering B of the same track)

 

[Jochen]

Probably not. What might create some correlation is people who care about higher fidelity are less likely to crush things with over-compression. They also are likely to have a wider technical dynamic range. Yet, it is quite possible to create a low DR recording with the highest dynamic range gear. One is not a requirement or directly linked to the other.

I'll introduce some simple complications (is that an oxymoron?). Many people have listening rooms that on a sound level meter would read around 40 db SPL for noise. You'll find it recommended that your gear should have perhaps 105 db SPL capability in smaller listening spaces. Does that mean 65 db dynamic range is all you could possibly hear?

No. For one reason we can hear into noise perhaps 15-20 db. Yet that varies with frequency. So is 85 db dynamic range the most we could hear? No, because noise has a pinkish to brownish quality. Most of the energy is at low frequencies we don't have much sensitivity to hearing. Our hearing kind of breaks sound into ERB filters (effective rectangular bandwidth). In the band with the greatest sensitivity around 3 or 4 khz a room with 40 dbSPL noise might only have noise of 10 dbSPL or less. That is also the area we can hear most deeply into noise. So in such a somewhat noisy environment in some parts of the spectrum we might hear right down to near 0 dbSPL. So you might need gear that can do 105 db of dynamic range. OTOH, if someone hits you with 105 db even for a moment during a transient it will be some period of time before you could hear the lowest level signals even if the gear and recording has them.

Now further, one can create synthetic music with 140 db of dynamic range. If however the source is from a microphone, guitar amp or such there are limitations on how quiet it can be from background noise, microphone self noise, and noise in the microphone preamp. While one might process it to create more it is very uncommon to manage something with more than 80 db of dynamic range. Further for most music most people don't want that. They prefer a bit of compression. You might say you don't, and there are times I don't. Music being a commercial venture means they will cater to what most people want or use. Also surprisingly, some tastefully done compression will be perceived as much more dynamic than uncompressed music which has the far higher measured dynamic range.

It is simple a mathematical/statistcal analysis of the signal/time series. That has nothing to do with listening rooms. If you have three quantities, A(maximum), B(minimum), and C (rms) and then build two new quantities D = A/B and E = A/C, and then calculate the correlation of D and E over various time series, they will be correlated.

 

[Blumlein 88]

It is simple a mathematical/statistcal analysis of the signal/time series. That has nothing to do with listening rooms. If you have three quantities, A(maximum), B(minimum), and C (rms) and then build two new quantities D = A/B and E = A/C, and then calculate the correlation of D and E over various time series, they will be correlated.

So now I no longer seem to be able to follow the plot. I'll bow out.

You post supposition is wrong however. For instance D can be exactly the same for multiple tracks and E be different for every one.

 

[solderdude]

Thus I would assume a quite high correlation between these two measures.

Nope... the difference might be in the order of 80 to 100dB and will still have no relation.
They are simply different 'dynamic range' concepts entirely.

 

[Adi777]

Nope... the difference might be in the order of 80 to 100dB and will still have no relation.
They are simply different 'dynamic range' concepts entirely.

Is vinyl due sometimes, often? higher DR can make it sound better than CD or streaming?

 

[danadam]

The building up at around second 3 or 4 is definitely not silence. So even if you cut the measurement region before that and after around second 58, the result will not change much.

If 25 dB isn't much. Here's the whole track:

]$ sox "Overwatch Music - (19) We Are Overwatch-nhwdxP6tlE0.flac" -n stats
             Overall     Left      Right
...
RMS Tr dB     -83.29    -83.24    -83.29
...
Window s       0.050

and here's with 1 second trimmed from the beginning and 5 seconds trimmed from the end:

]$ sox "Overwatch Music - (19) We Are Overwatch-nhwdxP6tlE0.flac" -n trim 1 -5 stats
             Overall     Left      Right
...
RMS Tr dB     -57.25    -57.25    -54.78
...
Window s       0.050

By the way, who says that this is silence ? It is very quiet, exactly because of this this recording has a high DR.

But you have no way of telling from the number if it was actual content that was quiet or if it was just fade in/out. Here is white noise with 5 dB:

]$ sox -n -n synth 60 white stats
...
RMS Tr dB      -4.97
...
Window s       0.050

Let's add 1 second of fade in and just like that we have 83 dB:

]$ sox -n -n synth 60 white fade 1 stats
...
RMS Tr dB     -83.11
...
Window s       0.050

 

[Newman]

Is vinyl due sometimes, often? higher DR can make it sound better than CD or streaming?

As long as you are not using that tool to form the opinion that vinyl has higher, um, DR. It lies horribly about vinyl, and can't be trusted for digital either.

Ears don't work for this task either. I once saw a recording engineer playing online one of his own recordings that he had ripped from digital and vinyl, which he knew for a fact had the same, for want of a better term, musical dynamics. Cue numerous comments from listeners that the vinyl definitely was more dynamic sounding, and he's lying. That's the power of just thinking it's vinyl and not brick walled, and seeing the higher DR-tool number, and seeing the more jagged-looking waveform.

So if you can't trust your ears, sighted, and you can't trust that tool, nor the visual waveform, how exactly do you plan to establish which is less dynamically compressed sounding?

 

[little-endian]

Many people have listening rooms that on a sound level meter would read around 40 db SPL for noise. You'll find it recommended that your gear should have perhaps 105 db SPL capability in smaller listening spaces. Does that mean 65 db dynamic range is all you could possibly hear?

No. For one reason we can hear into noise perhaps 15-20 db. Yet that varies with frequency. So is 85 db dynamic range the most we could hear? No, because noise has a pinkish to brownish quality.

That argument goes into a similar direction which the engineers from Benchmark made then they explained why they opted for a "prone-to-wear on the negative side but keeping the SNR at all positions on the positive" kind of potentiometer in the analog path to control the volume or rather "gain" with their DAC1, as normal listening levels a lot below 120dBSPL would still require a higher dynamic rage than assumed at first (peaking around your quoted 105dB or so).

Nowadays, that consideration should be obsolete entirely anyway to to even higher SNRs of DACs so some mathematical attenuation doesn't hurt anymore if it ever really did.

OTOH, if someone hits you with 105 db even for a moment during a transient it will be some period of time before you could hear the lowest level signals even if the gear and recording has them.

That is part of a possible counter-argument which makes me think how relevant all this really is in practise. If one gives human's hearing capabilities additional credit for achieving perception below noise floors (whose contradiction for me only is resolvable through the presumed actual misleading way we express noise floors to begin with - as a single number although it varies depending on the range), one should also substract quite a lot due to its shortcomings. As you say - the dynamic range even within such a narrow frequency band isn't always available since the cells in the ear need recovery, like the eyes despite quick adaptation to light. So the whole suggested dynamic range capability is quite optimistic and requires ideal environments.

On the other hand, to get to the theoretical ideals, of course on has to assume the best conditions, given.

The question for me now is, how we should state the dynamic range of an average (older) song whose peaks reach 0dBFS (lets take the intersample thing either out of the equation or add a few dB in the plus range) and whose level during the song in the middle never falls below -30dBFS or so.

Should be refer to that average like DR meter roughly does it or should be split the whole thing into many frequency bands and measure parts of that again? Depending on the frequencies, a lot higher values all of the sudden will be reached I guess.

But what really arrives in our consciousness after all the masking effects which especially lossy codecs (very successfully) rely on?

 

[solderdude]

Is vinyl due sometimes, often? higher DR can make it sound better than CD or streaming?

The differences in DR numbers (not necessarily DR) can stem from a different master to 'errors' and even RIAA correction errors or peaking of cartridge/pre-amp combinations.

 

[Sokel]

For anyone who's curious,the very top album of the list in TT is Denon's technical CD with DR 48 and specifically this:

It's not a song of course,most of this CD is test tones and that's an impulse.

 

[solderdude]

Fooling the TT meter's DR number generation with special tracks and test signals has no relevance to the S/N or THD+N/S ratio nor to any perception of dynamic range.