Upper Frequency in music - essential Pixie Dust or Digital Detritus?

QuadraphonicQuad

Help Support QuadraphonicQuad:

This site may earn a commission from merchant affiliate links, including eBay, Amazon, and others.

J. PUPSTER

💿🐕 Senior Disc Chaser 🎸
QQ Supporter
Joined
May 30, 2017
Messages
13,005
Location
CALIFORNIA (CENTRAL)
Since most of us can't hear above around 20K frequency range (and most of us even much lower than that) and with limitations of speakers etc.; are frequencies above that level as represented below in a Spectrogram, (DTS of On The Border "Year Of The Cat"-fronts only) important Pixie Dust that enhances the listening experience, or Digital Detritus that only creates distressing ultrasonic noise?

on the border spectrog.jpg
 
If you look at the frequency spectrum so as frequency versus amplitude rather than as a spectrogram (which is frequency versus time) what you would see would be the slope of the anti-alias low pass filter as you got above 20kHz, the vast majority of the 'signal' above 20k we can't hear and a lot would actually just be noise. By using a higher sampling frequency we can use Bessel filters, which have a slower roll-off but the best impulse response, and still be able to reduce the signal level so it isn't folded back in frequency by sampling (anything above half the sample rate is folded back, so we need to attenuate it to less than the resolution of the convertor). We can also spread the inherent quantisation noise due to digitally sampling over a wider frequency range which improves the stuff we do hear. There is a lot of mysticism that has been associated with higher sample rates it is not done to capture the un-hearable but to improve the hearable. So the stuff above 20kHz that actually does get reproduced only irritates dogs, cats & bats!
 
If you look at the frequency spectrum so as frequency versus amplitude rather than as a spectrogram (which is frequency versus time) what you would see would be the slope of the anti-alias low pass filter as you got above 20kHz, the vast majority of the 'signal' above 20k we can't hear and a lot would actually just be noise. By using a higher sampling frequency we can use Bessel filters, which have a slower roll-off but the best impulse response, and still be able to reduce the signal level so it isn't folded back in frequency by sampling (anything above half the sample rate is folded back, so we need to attenuate it to less than the resolution of the convertor). We can also spread the inherent quantisation noise due to digitally sampling over a wider frequency range which improves the stuff we do hear. There is a lot of mysticism that has been associated with higher sample rates it is not done to capture the un-hearable but to improve the hearable. So the stuff above 20kHz that actually does get reproduced only irritates dogs, cats & bats!
Uhh, sure, about as clear as mud I guess -LOL

I like pictures! Does this Freqy Chart represent what you're talking about Dunc?

Freqy.jpg
 
Why would the highest amplitude level be -27dB below max?
Don't know this stuff well at all, that's why I started this thread.

I believe there can be all sorts of stuff to use Specs for; as an example...
Below is a Spec. of an Up-mix I'm working on. This is the tail end of Train Train by Blackfoot from an LP rip I just did. I could hear the little scuff noises, but had a hard time seeing them on the wave forms. But if you look at it in the Spec chart, you can easily see where they are (Spec of Left Rear and Wave on Right Rear both with the vinyl scuff sounds.) I'd already ran this through RX 8 De-Click and De-Crackle, but it just won't take it out.

Also, thinking of ways this info can be utilized as a way to isolate certain freq. ranges to better Stem out instruments. I'm thinking that could be what's going on behind the software when DAWs like RX 8 create stems in their module "Music Rebalance."

SPEC FREQY LP SCUFFS.jpg


RX8 MR.jpg
 
Why would the highest amplitude level be -27dB below max?
In the time domain the composite signal (so all frequencies combined to produce a waveform in time) will be above that, that is the signal the Analogue to Digital Convertor 'sees' and when converting to the frequency domain the graph displays the relative levels that constitute the waveform(there may be some scaling on the spectrum amplitude axis though).
 
Don't know this stuff well at all, that's why I started this thread.

I believe there can be all sorts of stuff to use Specs for; as an example...
Below is a Spec. of an Up-mix I'm working on. This is the tail end of Train Train by Blackfoot from an LP rip I just did. I could hear the little scuff noises, but had a hard time seeing them on the wave forms. But if you look at it in the Spec chart, you can easily see where they are (Spec of Left Rear and Wave on Right Rear both with the vinyl scuff sounds.) I'd already ran this through RX 8 De-Click and De-Crackle, but it just won't take it out.

Also, thinking of ways this info can be utilized as a way to isolate certain freq. ranges to better Stem out instruments. I'm thinking that could be what's going on behind the software when DAWs like RX 8 create stems in their module "Music Rebalance."

View attachment 65921

View attachment 65922
Isolating signals in time and frequency range is how I think instruments were isolated for albums like the surround mix of Miles Davis' Pictures Of Spain (at least I think that was the one I was thinking of!!)
 
If you mean the extended frequency range you get in HD sample rates (88.2k and above), any data there is an artifact. The entire point is to give the actual audio band (20-20k) a VERY wide margin.

If the sampling frequency is right at the edge of the data band you need to put a steep filter there. This is a difficult analog circuit that's part of a converter circuit. THAT is actually the part you're critiquing between different converter units. How well the analog stages of the converter circuits are dialed in and how good those low pass filters are in SD mode. HD avoids that with wide margins. They all use mostly the same digitizer chips. Sometimes the simple solution is the way to go!

And you really can upsample SD program if your DAC unit sounds better at HD. :)

The more boutique DAC units pretty much sound just as good at SD as HD. That's part of what you paid for.

Related:
I believe 96k/24 bit to be a complete container for audio no matter what's happened to that audio previously.

I wanted to see if 192k held on to something extra.
You know how people have uploaded a video to Youtube 100 times to exaggerate the compression artifacts? The point of that is to show they are there even though you might not notice them with one pass.

So I did that with 192k vs 96k
Source: A side of Dave Crosby - Remember My Name 45 rpm Classic Records vinyl
Apogee Rosetta 800 192k
A high fidelity source.

192k original -> 96k -> 192k

This iteration #3 here is the first return to 192k.
Obviously there's a difference with whatever ultrasonic artifacts truncated. Sounds identical to me and nulls down to 90db or so on the meter. The difference signal is not audible in any way even with dangerous monitor levels.

On we go until iteration #100 back to 192k.
#3 and #100 null perfectly to digital zero. The audio band has not been touched. Not even 1 bit changed over all those conversions.

So I like 96k 24 bit digital. It just works no matter what. And works the best in all AD and DA converters no matter what.
 
Part of the stuff above 20 KHz is alias artifact from the digitization of the signal.

Anything at or above the Nyquist frequency is garbage due to aliasing.
 
Part of the stuff above 20 KHz is alias artifact from the digitization of the signal.

Anything at or above the Nyquist frequency is garbage due to aliasing.
Just to be clear. That's when things are going wrong!

Aliasing isn't an acceptable error in any way! This is to be avoided at all costs. Sounds bad. Literally the entire reason for making and dealing with difficult low pass filters in the analog stages of the converter circuits! The older sample rate conversion algorithms had problems with aliasing too and that's why those were avoided in the early days.
 
So there is nothing on the recording above the sample rate that is usable material.

If you are seeing anything above that frequency on a spectrogram, it is caused by false harmonics of material below that frequency being misread by the spectrogram program.
 
From a technical perspective the Spectrogram isn't really of much use to man nor beast, its just a visualisation, a 'heat map' that uses colour to indicate things are happening over different frequencies with time, so I wouldn't use it to assess anything other than to show a nice picture. The FFT spectrum produced by the FFT (Fast Fourier Transformation) gives technically useful data so allows you to measure the signal level of a particular frequency.
 
From a technical perspective the Spectrogram isn't really of much use to man nor beast, its just a visualisation, a 'heat map' that uses colour to indicate things are happening over different frequencies with time, so I wouldn't use it to assess anything other than to show a nice picture. The FFT spectrum produced by the FFT (Fast Fourier Transformation) gives technically useful data so allows you to measure the signal level of a particular frequency.
What's that look like?
 
Were you aware that the cochleas in our ears do mechanical Fourier transforms so we can hear pitch?
 
Were you aware that the cochleas in our ears do mechanical Fourier transforms so we can hear pitch?


Translation / time shifting

Animation showing the Fourier Transform of a time shifted signal. [Top] the original signal (yellow), is continuously time shifted (blue). [Bottom] The resultant Fourier Transform of the time shifted signal. Note how the higher frequency components revolve in complex plane faster than the lower frequency components.

Had to look it up 😎
 
I struggled in college just to pull a B+ out of my Calculus class. And I don’t believe I’ve used much, if any of that excruciating experience since then. But teachers (good teachers) are my heroes. They put you in a place of understanding the full picture. One can learn the symbols and rules of Calculus; but unless you fully grasp how it applies to the physical mechanics of the real world, it is useless.
 
I struggled in college just to pull a B+ out of my Calculus class. And I don’t believe I’ve used much, if any of that excruciating experience since then. But teachers (good teachers) are my heroes. They put you in a place of understanding the full picture. One can learn the symbols and rules of Calculus; but unless you fully grasp how it applies to the physical mechanics of the real world, it is useless.

Well, that plot is not calculus... the Fourier Transform plot is showing how there are components at the fundamental frequency, the first odd ( 3f ) and the second odd (5f) harmonics. And also how the real and imaginary components are out of phase ( I think by 90 degrees.. a cosine, I think.. notice the rotation phase angles when the real is max and the imaginary is 0... and so on...). Plus the fact that the harmonics rotate faster than the fundamental.

I don't know if this is calculus at all. I learned it in Physics -my degree.
 
Back
Top