UnderTow


http://www.youtube.com/watch?v=XyOHJa5Vj5Y&fmt=18

UnderTow

Interesting!

Well, I was hoping someone would mention the Nyquist theorem, so here it is, in my own words:

In order to sample something at frequency A, the sampling rate must be 2A.

Now, since the upper frequency limit of human hearing is about 20kHz, you need at least a sampling rate of 40kHz. Just to make sure you get those really high highs, you've got your 44.1. And just to make it a cleaner recording, you've got your 48, which is what I use. Using a sampling rate higher than 48, depending on your system, may sound better because of how your hardware works, and it may push the noise floor down during recording. That's about it. Theoretically, above 48 is not going to make it amazingly better. If it does, then your hardware isn't doing what it should. But if you're stuck with that hardware, then by all means use the settings that give you the best sound.

Dynamic range is handled by your bit rate, and 24 is what you want for the best recordings.

It's really all about the hardware you have and what you're trying to do. In an ideal world, where all s h i t does what it's supposed to do, for music recording you should never need above 48/24.

Needless to say, our little society on this planet is far from ideal. And so, class, you use what works best in your particular rig.

(P.S. Come on, we're not preschool kids here. Is it really necessary to filter out the word s h i t?)

wintaper


I fully understand digital audio and nyquist. You're missing my point - no one is arguing the math (well not me anyway) . There's is a lot more to digital audio than just nyquist.

a higher sample rate (and bit depth) can produce a more accurate waveform - by reducing the margin of error. Whether this is audible - or worth it - is a completely different matter.

Lets use a digital photocopier as an analogy.

*sigh*  You can't compare visual stuff with sound.  It doesn't work the same way at all. Please read the papers linked in this thread. Like this one:

http://www.lavryengineeri...ts/Sampling_Theory.pdf

This paper is written by Dan Lavry. He is one of the most respected converter manufacturers. He fully understands the theory, the maths and their implementation in the real world.

UnderTow

lol - this is becoming one of those "two years later.." threads ...

ill take a look here again when we hit postnumber 300 - because i think that 300 might be an adequate resolution (ups sorry - samplerate) to reveal an average value of truth that should be representable in the anal worls (did i say "anal"??? - well i mean this purely from the perspective of psychoanalisis - nothing sexual...)

btw i built a trompnone with an included compressor that has a dynamic range of 72 bit - can i record it with a soundblaster ??? (the name "blaster" would allude to this)

good- byyyyyyye

BTW what is the opposite of a digital pictures? Is there such a thing as an analog picture?

"analog" would actually be continuous as in a photo - no 'in-between areas" like newspapers and magazines have.


the analogy was simply meant to indicate that the copy with more colors would generally look better than one with less and nothing else.

wintaper

BTW what is the opposite of a digital pictures? Is there such a thing as an analog picture?

"analog" would actually be continuous as in a photo - no 'in-between areas" like newspapers and magazines have.


the analogy was simply meant to indicate that the copy with more colors would generally look better than one with less and nothing else.

 
The range of color comes from the bitness of the file, not from its pixels.
 
 
 

Okay. Since the "BSers" seem to be so fond of analogies, how about one for "our" side:

A circle in two-dimensional space can be completely defined by giving the coordinates of its center, and the scalar value of its radius. It is not necessary to record the coordinates of an infinite number of points on the circle to accurately reproduce it. All you need is the mathematical rule for generating a circle from this information, which is (x-h)² + (y-k)²=r², where h and k are the x and y coordinates of the center of a circle radius r.

Waveform sampling works similarly. If you have two data points for each cycle of the highest frequency you wish to record and reproduce, those data points completely define the shape of a waveform containing any and all frequency components up to and including that frequency limit. No additional information is necessary to perfectly reproduce the analog waveform from which those samples were taken. The key is that you don't just "connect the dots", you have to know and apply the rule for mathematically reconstructing the waveform represented by those data.

If it does, then your hardware isn't doing what it should. But if you're stuck with that hardware, then by all means use the settings that give you the best sound.

 
Yes ... again to keep this on topic. It is possible that Freddie is hearing an improvement, due to HIS hardware. Even the lavry article points out that different converters are optimized for different sample rates. So in Freddie's case, his converters might be adding less distortion at 96k.
 
Dude

Wintaper ..... yes .. I know overtones ... and harmonics .... etc. But I (nor you nor anyone) can hear any of them above around 20K - that's the point.

First point - It is absolutely true that for "music" there are frequencies that go above 20K - in fact they can go on forever. It doesn't matter - you can't hear them. They do not need to be captured (with the possible caveats a couple of us have pointed out)

Second point - the photo resolution analogy is not valid - not the same concept at all. I'll try it real simple.

If you have sampled at the Nyquist Frequency, you have all the data needed to construct the waveform completely and perfectly and more data won't increase that "resolution"!

brundlefly


Okay. Since the "BSers" seem to be so fond of analogies, how about one for "our" side:

A circle in two-dimensional space can be completely defined by giving the coordinates of its center, and the scalar value of its radius. It is not necessary to record the coordinates of an infinite number of points on the circle to accurately reproduce it. All you need is the mathematical rule for generating a circle from this information, which is (x-h)² + (y-k)²=r², where h and k are the x and y coordinates of the center of a circle radius r.

Waveform sampling works similarly. If you have two data points for each cycle of the highest frequency you wish to record and reproduce, those data points completely define the shape of a waveform containing any and all frequency components up to and including that frequency limit. No additional information is necessary to perfectly reproduce the analog waveform from which those samples were taken. The key is that you don't just "connect the dots", you have to know and apply the rule for mathematically reconstructing the waveform represented by those data.

good analogy - very well said - I like that!

wintaper


most people with trained ears can tell a sine wave from a square wave ..

Only waves up to a certain frequency as pointed out - if the harmonic that makes it sound "square" isn't there - you don't hear it as square.

wintaper

The range of color comes from the bitness of the file, not from its pixels

um - did I say pixels - or did yo say that for me? cause I don't see that in my post and sure as hell wouldn't have said it. Please don't make things up in my part.

 
My bad. 
 
When you said: "analog" would actually be continuous as in a photo - no 'in-between areas" like newspapers and magazines have."
 
I thought you were talking about pixels, which is why I replied with:
 
"The range of color comes from the bitness of the file, not from its pixels"
 
I apologize for the confusion.
 
 
 

wintaper


Frequencies above the audible range affect frequencies we can hear. The problem of missing overtones in high frequency content like cymbals caused a lot of the "harshness" associated with early A/D converters (due to poor low-pass filters before the conversion process). Modern over-sampling has pretty much solved this problem - as pointed out in an earlier post. But those frequencies do matter.

well - now you are into the argument that I acknowledged has a lot of controversy around it. Many reasonable people say this is nuts - Many reasonable people agree with you. I have not made up my mind on it personally. Given that MOST adults can't even hear past 16 or 17 K - even 44.1 gives a lot of info they don't need.

But this is a different argument.

It is one thing to say that the higher frequency content is important and affect how we experience sound. It is quite another to try to say that for a given frequency, anything greater than the N-freq adds any value - and that's what several (including you as a I remember) have seemed to try to say here!

wintaper


most people with trained ears can tell a sine wave from a square wave - or sawtooth for that matter - you don't need supersonic hearing. You just need to know what to listen for. 

Not a 12 Khz "square" wave. You can not tell the difference. A 5 Khz square "wave", sure.  Any of the "square" waves played by an analogue synth, sure. Not a 12 Khz "square" wave. You can not tell the difference.


UnderTow

for a given frequency, anything greater than the N-freq adds any value

don't think I ever said that - which is not to say someone might have interpreted it that way.


I DO believe frequencies outside the audible range affect what we hear, so it would be nice to sample at a high enough frequency to capture this. That being said - at the moment I do not hear a large enough difference to balance the negative aspects of 96kHz - for MY studio.