[castleofargh]

 

  1) Any REAL musical sound has two audible parts: The main and first one is a (more or less) periodic function of any form (Let's call it A) and a second one: The natural instrument noise (Not ambient noise), a non periodic function, so without any wave form, just pressure changes following not known rule (Let's call it part B). And is important to say here that, nevertheless, this noise has "its own unknown" rules; for instance, it is different at different frecuencies of part A.

I know this is old at this point, but I wanted to address this...
 
If a part of the music cannot be broken down into periodic frequency components, that means that it is effectively DC. You can't hear this, nor is it reproducible by speakers. Therefore, it doesn't matter. A randomly varying signal can still be described based on its frequency content, and just because you can't see a simple, repeating pattern doesn't mean the signal doesn't contain any frequencies.

yup he can get an idea of what we're talking about with something like this


 the more frequencies the less it looks like it can be made of waves, but it always is.

 

[efeuvete]

  So you can't hear music quality?
 
All ABX does is it presents you with two files and asks you to discern the difference by listening. There's no limitation on how long you can listen for it to be a valid test, there's no limitation on what you should be listening for (be it ambience, soundstage, stereo image, or weird artifacts in a particular flute solo). The only requirement is that you listen to the music, and see if you can tell the difference.
 
I find it very strange that people will wax poetically about how "musical" something sounds with a new component/format/etc, and how the change is night and day, and audible with any remotely decent equipment, but when asked a question as simple as "Pick the musical one out of these two samples, and do it consistently", suddenly the difference is so subtle that it can't be done? It doesn't stand up to scrutiny.

 
Hi, I must say again that If I write about this topic here is because I think it is an interesting one but, please, the fact that my opinion is obviously different of the main forum opinion, does not mean that I have any intention of disturbing anybody with it in any way. Once this is said, I go:
 
cjl: As I've said before, I'm not a math expert at all, so I may be wrong; But as I understand the Fourier Transform, what this math device can do is to give a "Frequency domain" interpretation of a sampled sound but, that does not mean that its interpretation is always an exact reproduction of any kind of sample.
 
I say this because following what a Fourier Transform (FT) is, what I can see is a relevant mathematical device which represents PERIODIC functions of any form as sums of infinite senoidal functions. So, even being those sums infinite, as the sum terms decrease in weight, is obvious to me that the FT will do a good job analyzing musical sounds; Even to the point of giving us a minimun level of sampling frequency via the Nyquist Theorem. But...
 
Then came the idea of applying this FT to any sequence of numbers which, as it seems, as a lot af math and physical applications as FT provides (almost always) a representation of that sequence of numbers as a sum of infinite weighted frecuencies (Frecuency domain). Well, no doubt this can be very useful (Depending of what the original sequence can mean) but, as far as I can understand, I'm not sure at all about "how precise is" the FT of a sampled thunderstorm noise or of a sampled slammed door noise (Natural NO PERIODIC noises).
Surely, the aproximation would be significative, and useful too, but I see no way to know how big is the difference between the FT of the sample of a natural noise and the sample itself, nor the error I can have applying Nyquist theorem to that sample. And, being the topic here the very subtle difference between 16/44100 and 24/96000, I' not sure at all "my" musical instrument noise does not get lost in the FT game.
 
Maybe here on the forum there's people who know maths enough (Or know sombebody who knows maths enough) to tell me if my opinion is wrong and why and, bellieve me, I would be as happy to fundament it as to discard it.
 
PS/ Castleofargh: I'm not saying that noise is not perceived as a sum of waves (It well can be, I don't know), what I'm saying is that I don't know if FT, and therefore, Nyquist theorem, is applicable samplig natural noise.   

 

[efeuvete]

  I know this is old at this point, but I wanted to address this...
 
If a part of the music cannot be broken down into periodic frequency components, that means that it is effectively DC. You can't hear this, nor is it reproducible by speakers. Therefore, it doesn't matter. A randomly varying signal can still be described based on its frequency content, and just because you can't see a simple, repeating pattern doesn't mean the signal doesn't contain any frequencies.

Sorry, I've quoted wrong. This one is the post I should have quoted.

 

[RRod]

 
Sorry, I've quoted wrong. This one is the post I should have quoted.

 
The requirement for the Sampling Theorem is that the signal must be bandlimited as I mentioned previously. If a signal is bandlimited, then the samples can exactly reproduce the signal (the proof of the theorem gives a theoretical reconstruction via sinc functions). But we can't bandlimit perfectly since our signals are finite in length, but we have ways of reducing this issue.
 
Whatever complex sound you think of first has to go through the microphone, which already has its own frequency limits. It then has to go through the ADC, which first hits it with an anti-aliasing filter so that the signal becomes approximately bandlimited. It then gets sampled. This is true at 24/96 just as much as at 16/44.1. The difference is the frequencies that get through, not magic non-frequency material.

 

[efeuvete]

I've posted twice instead of editing a detail. Sorry

 

[efeuvete]

   
The requirement for the Sampling Theorem is that the signal must be bandlimited as I mentioned previously. If a signal is bandlimited, then the samples can exactly reproduce the signal (the proof of the theorem gives a theoretical reconstruction via sinc functions). But we can't bandlimit perfectly since our signals are finite in length, but we have ways of reducing this issue.
 
Whatever complex sound you think of first has to go through the microphone, which already has its own frequency limits. It then has to go through the ADC, which first hits it with an anti-aliasing filter so that the signal becomes approximately bandlimited. It then gets sampled. This is true at 24/96 just as much as at 16/44.1. The difference is the frequencies that get through, not magic non-frequency material.

RRod, I've read your post, sorry I did not mention it but as I can not say up to what point "we have ways of reducing this issue" I didn't want to look like as critizicing somethiing which I do not know deeply enough.

 

[bigshot]

Within the range of human hearing (dynamics and frequency range), CD quality sound and high bitrate sound are bit for bit identical. High bitrate files don't contain any additional information that humans can hear. They contain information that is beyond the frequencies that ears can hear and below the threshold where things are too quiet to hear.

 

[efeuvete]

Maybe now you're going to wish killing me (If not already before) because I'm going to give you a "magical" example to synthetize how I see this "FT thing":
Imagine I use a telephone guide as a source of some thousand numbers and I reduce them to the range -32768 +32768. Then, if I apply a Fourier Transform to those numbers, I will surely get a "Frequency Domain" image of the Guide.
 
¿Does that FT mean that telephone numbers follow frequency patterns?
 
I'll log out for a while, just in case.

 

[cjl]

Interestingly, yes, you can extract frequency information from a set of numbers like that, and in some cases, it can even be meaningful (depending on how you arranged them in the first place).

 

[StanD]

  Maybe now you're going to wish killing me (If not already before) because I'm going to give you a "magical" example to synthetize how I see this "FT thing":
Imagine I use a telephone guide as a source of some thousand numbers and I reduce them to the range -32768 +32768. Then, if I apply a Fourier Transform to those numbers, I will surely get a "Frequency Domain" image of the Guide.
 
¿Does that FT mean that telephone numbers follow frequency patterns?
 
I'll log out for a while, just in case.

While you're in hiding you can go square and study up on Walsh Analysis for some fun.

 

[efeuvete]

  While you're in hiding you can go square and study up on Walsh Analysis for some fun.

 
(Mr, or Mrs) StanD, I'm afraid I have the same reasons to study that "Walsh Analysis" than you for studying concert piano. Anyhow, if you try studying Karl Popper, maybe you'll have even some more. That guy taught me to, orderly and respectfully, doubt.

 

[StanD]

   
(Mr, or Mrs) StanD, I'm afraid I have the same reasons to study that "Walsh Analysis" than you for studying concert piano. Anyhow, if you try studying Karl Popper, maybe you'll have even some more. That guy taught me to, orderly and respectfully, doubt.

This forum is about Sound Science, not concert piano. I did study a lot of music theory, Diatonic and Chromatic Harmony, Counterpoint, Orchestration and so on. None of that helps with the bits and bytes of sound reproduction,

 

[efeuvete]

But be trained to doubt is good for sound science too (As "my sound science case", win or lost, may end showing).

 

[RRod]

  Maybe now you're going to wish killing me (If not already before) because I'm going to give you a "magical" example to synthetize how I see this "FT thing":
Imagine I use a telephone guide as a source of some thousand numbers and I reduce them to the range -32768 +32768. Then, if I apply a Fourier Transform to those numbers, I will surely get a "Frequency Domain" image of the Guide.
 
¿Does that FT mean that telephone numbers follow frequency patterns?
 
I'll log out for a while, just in case.

 
Here, read this and then we can talk more about the FT, FS, DTFT, and DFT:
http://www.dspguide.com/pdfbook.htm
 
In a recording/playback chain without any digital filtering you don't even need to do any transforms. There it's all about the *analog* anti-aliasing and anti-imaging filters, and the electrical characteristics of the ADC, DAC, and transducers.

 

[bigshot]

It's interesting that someone would be well versed in the minutia of digital audio, but so oblivious to the basics.