Are microphones have enough sensitive enough for high-res audio?

[Petri Kaipiainen]

The major problem with this so called High Defenition recording is the fact that is is possible to achieve about 22 bit resolution with best mics, and frequency extension to above 50 kHz, but never with the same mic!!!

It is possible to have quitet mics only with large diaphrams, but high frequency extension only with small diaphram mics. These requirements exclude each other out. Thus true high defenition is just a dream, having high bit rate and sampling rate does not do any good if mics and analog components can not deliver, and they can not deliver even 24/96 quality.

Sanken super mic CO-100K: noise level: 22dB-A, max sound pressure 125 dB. Basically 17 bit performance in dynamic range, CD can achieve this with proper dithering.

DPA large diaphram 130 volt mic 4041-S: noise level 7dB-A, max sound pressure 144 dB. Basically 22 bit performance. But frequency range only extends to 20 kHz.

So there you are, no matter what marketing men try to sell, real high extension and low noise is impossible to achieve at the same time with microphones, only by synthesis.

[Steve Oakley]

Originally Posted by Petri Kaipiainen

Not so.

Sound is made up form a bunch of sine waves and the sum of these forms the complicated composite wave we can see in an audio editor. All such complex waveforms can be reduced to a set of sine waves (fourier series). The fact is humans can hear only to a certain frequency (about 20 kHz) and capturing and reproducing higher components can not be heard. The beauty of the sampling theory is that all waveforms below half of the sampling rate can be reproduced PERFECTLY. Having more samples than necessary DOES NOT improve the quality of those signals, only captures also higher frequencies. Which in this case in not necessary, as those components lay above our hearing limits.

The view you brought up is a common fallacy and not true.

ok, if the reproduction was perfect, why can I hear the difference between 44.1 and 48 K ? and certainly 96k ? I think even a "tin"ear could pick out 32K easily enough. the problem between theory and practice is reality. if one where to model sound in terms of pure sine waves, they you lose other transients which are not curved. you would be rounding them off and creating a less accurate sample of them. the more samples, the more accurately the wave from is sampled, even if there is some mathematical cheating of rounding. otherwise no one would be able to hear the difference between sample rates, and clearly many people can. what people lose site of is that 44.1 is a number from the 1960's. back then digital technology was pretty limited and 44.1kz was considered the threshold of where some one started to have a problem between live and source, but it was not the ideal point, just the minimum. if they could of drove faster sample rates back then, they would of but the technology became too difficult or expensive.

[Petri Kaipiainen]

Have you ever tried a blind test where the only difference in the signals is the sample rate (high cutoff frequency)? If you really can hear reliably a difference of 44, 48 and 96 you are a truly exeptional person. Most of us can not hear much above 16 kHz.

We do loose those transients which are too sharp, true, but those transients contain frequencies which are outside of human hearing. And even those "which are not curved" are composed of even higher sine waves... But the crux of the matter is that those frequencies are not heard, thus they need not be recorded! There is nothing lost in what we hear.

You say "many people can [hear the difference between sample rates]", but how come in the AES test out of 100 test subjects NONE was able to tell the difference between SACD and CD? Wrong people chosen for the test, again, why did they not ask you or many other of those who claim to hear the difference? Maybe nobody can hear the difference in real life, after all?

Try the test file at http://hosted.filefront.com/Jullepoika/ to see if you really can tell the difference between 16/44 and 24/96. You only need a true 24/96 capable system to test it.

[A. J. deLange]

Originally Posted by Steve Oakley

ok, if the reproduction was perfect, why can I hear the difference between 44.1 and 48 K ? and certainly 96k ?

The reproduction isn't perfect but it might as well be.

There are 2 possibilities as to why you may hear, or think you hear differences. One, the more likely these days, is that that marvellous piece of equipment between your ears is equipped with a special processing unit called the "imagination". It is extremely powerful device capable of permitting you to hear what you expect to hear even when it isn't there. To say that you can hear a difference is only meaningful if you did the test with an A,B,C box in which two of the buttons are wired to one signal and the third to the other. You cannot know which is which and ideally the person conducting the test shouldn't either. This is called a double blind triangle test. You must pick which two of the buttons are wired to the same signal and determine which sounds better and you must do this on enough different sound samples that the result has statistical significance. When triangle tests are performed the ability to hear the distinctions that are widely reported on in the popular audiophile press disappear. That's why you don't see much about them. A colleague mentioned to me the other day that he sometimes reads a forum where the rules have only one prohibition: mention of triangle tests.

The other possibility is that the low sampling rate system was not properly implemented. In today's world that's not likely but it certainly has happened in the past and will probably continue to do so. Most OEMs use A/D converters in which proper DSP is built in. The architecture is usually sigma-delta in which at most a few bits (1-4) are sampled at a rate of several MHz and these samples filtered and decimated to provide the outputs of up to 24 bits. The manufacturers test their designs extensively and report test data to the OEMs who integrate the chips into the stuff we buy. You'd have to work pretty hard at it to screw up with one of these devices but I suppose it can be done.

The sampling theorem (note: it is a theorem - it has a proof) requires that signals be "strictly band limited". This requires that they be of infinite duration. It should thus, be obvious that there are no devices which adhere exactly to the sampling theorem because in the real world there are no strictly band limited signals. Nonetheless, for all practical purposes the sampling theorem does apply. The "errors" (quantizing noise, aliasing) can be controlled with proper DSP and pushed down to better than 100 dB below the signal (1 microvolt of noise re a 1 volt signal is 120 dB). I won't say there is no one that can perceive that but 999 out of as thousand who think they can are fooling themselves.

[Yi Fong Yu]

so, to sum it up:
1. mics can't capture more than what humans can hear.
2. humans can't heart worth a damn anyways
3. capture @the highest quality you can so when you render after final edit, the quality is maintained.

very interesting =D. thanks ya'll =D